Reactive Simulation of Shoring and Excavation Processes Based On Automated Performance Monitoring Maximilian Bügler

Ingenieurfakultät Bau Geo Umwelt
Lehrstuhl für Computergestützte Modellierung und Simulation
Reactive Simulation of Shoring and Excavation Processes based on

Automated Performance Monitoring
Maximilian Bügler
Vollständiger Abdruck der von der Ingenieurfakultät Bau Geo Umwelt der
Technischen Universität München zur Erlangung des akademischen Grades
eines
Doktor-Ingenieurs
genehmigten Dissertation.
Vorsitzender: Univ.-Prof. Dr. rer. nat. Ernst Rank

Prüfer der Dissertation: 1. Univ.-Prof. Dr.-Ing. André Borrmann
2. Univ.-Prof. Dr.-Ing. Markus König
Ruhr-Universität Bochum
Die Dissertation wurde am 22.06.2016 bei der Technischen Universität

München eingereicht und durch die Ingenieurfakultät Bau Geo Umwelt am
17.10.2016 angenommen.
3
Abstract
This Thesis presents novel concepts for reactive simulation of shoring and excavation
processes. In this context methods for construction schedule optimization are combined
with techniques for real-time data acquisition. Furthermore uncertainty in the durations of
activities and the performance factors of involved resources are considered. This Thesis
presents a new schedule generation scheme, two heuristic scheduling algorithms, and a
concept to automatically track excavation processes.
5
Acknowledgements
First and foremost, I would like to express my sincere gratitude to my advisor Prof. Dr.-
Ing. André Borrmann for the continuous support of my research, for his patience, knowl-
edge, and motivation. His guidance helped me throughout the four years of my research
and the writing of this Thesis. I could not have imagined a better advisor for my Ph.D
study.
Besides my advisor, I would like to thank the rest of my Thesis committee: Prof.
Dr.rer.nat Ernst Rank, and Prof. Dr.-Ing. Markus König, for their insightful comments
and encouragement.
My sincere thanks also goes to Prof. Dr. Daniel Straub, Prof. Patricio Antonio Vela,
Prof. Dr.-Ing. Dipl.-Wi.-Ing. Willibald A. Günthner, and Prof. Dr.-Ing. Christoph van
Treeck, who gave me opportunities to collaborate in joint research endeavors.
I thank the Bavarian Research Foundation for funding the FAUST project which I
worked on throughout my study.
Furthermore I thank my fellow colleagues for the stimulating discussions, the hard
work on joint publications, the unforgettable experiences we had during conference trav-
els, and the amazing four years we have spent together.
Last but not the least, I would like to thank my family: my parents and my brother for
supporting me throughout writing this Thesis.
Contents 7
Contents
1 Introduction 11
1.1 Problem Statement and Scope . . . . . . . . . . . . . . . . . . . . . . . 12
1.2 Hypotheses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.3 Methodological Approach . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.4 Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.5 Terminology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2 State-of-the-Art in Modeling and Simulation of Construction Operations 18

2.1 Performance Factors . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.2 Modeling Uncertainty in Construction Operations . . . . . . . . . . . . . 20
2.2.1 Beta Distribution . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.2.2 Normal Distribution . . . . . . . . . . . . . . . . . . . . . . . . 21
2.2.3 Gamma Distribution . . . . . . . . . . . . . . . . . . . . . . . . 23
2.2.4 Weibull Distribution . . . . . . . . . . . . . . . . . . . . . . . . 24
2.2.5 Goodness-of-fit Tests . . . . . . . . . . . . . . . . . . . . . . . . 25
2.2.6 Monte-Carlo-Simulation . . . . . . . . . . . . . . . . . . . . . . 27
2.2.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.3 Modeling Costs of Construction Operations . . . . . . . . . . . . . . . . 28
2.4 Granularity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.5 Modeling and Simulation of Construction Processes . . . . . . . . . . . . 30
2.5.1 The Bored Pile Wall Example . . . . . . . . . . . . . . . . . . . 30
2.5.2 General Purpose Systems Simulation (GPSS) . . . . . . . . . . . 30
2.5.3 Activity Cycle Diagram (ACD) . . . . . . . . . . . . . . . . . . 34
2.5.4 CYCLic Operations NEtwork (CYCLONE) . . . . . . . . . . . . 35
2.5.5 STate and ResOurce Based Simulation of COnstruction ProcEsses
(STROBOSCOPE) . . . . . . . . . . . . . . . . . . . . . . . . . 37
2.5.6 Activity-Based Construction (ABC) . . . . . . . . . . . . . . . . 38
2.6 Modeling and Simulation of Construction Projects . . . . . . . . . . . . . 39
2.6.1 Activities-on-the-Arrows (AoA) versus Activities-on-the-Nodes
(AoN) representation . . . . . . . . . . . . . . . . . . . . . . . . 41
2.6.2 Elements of a Construction Project Model . . . . . . . . . . . . . 42
2.6.2.1 Activities . . . . . . . . . . . . . . . . . . . . . . . . . 42
2.6.2.2 Resources . . . . . . . . . . . . . . . . . . . . . . . . 42
2.6.2.3 Materials . . . . . . . . . . . . . . . . . . . . . . . . . 43
2.6.2.4 Geometric Space . . . . . . . . . . . . . . . . . . . . . 43
8 Contents
2.6.2.5 Precedence Constraints . . . . . . . . . . . . . . . . . 43

2.6.3 Computer-Aided Design (CAD) and Building Information Mod-
eling (BIM) as Data Source . . . . . . . . . . . . . . . . . . . . 43
2.6.4 Discrete-Event Simulation of Construction Projects . . . . . . . . 46
2.7 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
3 State-of-the-Art in Construction Schedule Optimization 50

3.1 The Anchored Bored Pile Wall Project Example . . . . . . . . . . . . . . 51
3.2 Objectives of Optimization and Schedule Metrics . . . . . . . . . . . . . 51
3.2.1 Uncertainty and Robustness . . . . . . . . . . . . . . . . . . . . 52
3.2.2 Multi-Criteria Optimization . . . . . . . . . . . . . . . . . . . . 53
3.2.2.1 Pareto Optimality . . . . . . . . . . . . . . . . . . . . 54
3.2.2.2 Compromise Solution . . . . . . . . . . . . . . . . . . 55
3.2.2.3 Weighted Global Criterion . . . . . . . . . . . . . . . . 55
3.2.2.4 Weighted Sum . . . . . . . . . . . . . . . . . . . . . . 56
3.3 Project Scheduling Problem . . . . . . . . . . . . . . . . . . . . . . . . 56
3.3.1 Precedence Constraints . . . . . . . . . . . . . . . . . . . . . . . 56
3.3.2 Graphical Representation of the Bored Pile Wall Project Example 57
3.3.3 Critical Path Method (CPM) . . . . . . . . . . . . . . . . . . . . 58
3.3.4 Program Evaluation and Review Technique (PERT) . . . . . . . . 62
3.4 Resource Constrained Project Scheduling Problem . . . . . . . . . . . . 65
3.4.1 Graphical Representation . . . . . . . . . . . . . . . . . . . . . . 65
3.4.2 Problem Characteristics . . . . . . . . . . . . . . . . . . . . . . 66
3.4.3 The Anchored Bored Pile Wall Project Example . . . . . . . . . 68
3.4.4 Schedule Generation Schemes . . . . . . . . . . . . . . . . . . . 71
3.4.4.1 Types of schedules . . . . . . . . . . . . . . . . . . . . 71
3.4.4.2 Serial Schedule Generation Schemes (SSGS) . . . . . . 73
3.4.4.3 Parallel Schedule Generation Scheme (PSGS) . . . . . 76
3.4.4.4 Differences between the Schedule Generation Schemes 78
3.4.5 Scheduling using Simulation . . . . . . . . . . . . . . . . . . . . 79
3.4.6 Simulation-based Critical Path Analysis using the Dori-Algorithm 81
3.4.7 Deterministic Approaches to the Resource Constrained Project
Scheduling Problem . . . . . . . . . . . . . . . . . . . . . . . . 84
3.4.8 Heuristic Approaches to the Resource Constrained Project Schedul-
ing Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
3.4.8.1 Priority Rules . . . . . . . . . . . . . . . . . . . . . . 86
3.4.8.2 X-Pass Methods . . . . . . . . . . . . . . . . . . . . . 86
3.4.8.3 Tabu Search . . . . . . . . . . . . . . . . . . . . . . . 87
3.4.8.4 Simulated Annealing . . . . . . . . . . . . . . . . . . . 87
3.4.8.5 Genetic Algorithms . . . . . . . . . . . . . . . . . . . 88
3.4.8.6 Ant Colony Optimization . . . . . . . . . . . . . . . . 88
3.4.8.7 Local Search Procedures . . . . . . . . . . . . . . . . . 88
3.4.8.8 Swap-based Local Search . . . . . . . . . . . . . . . . 88
3.4.8.9 Overview of Heuristic Procedures for RCPSPs . . . . . 94
3.4.9 Benchmarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
3.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
Contents 9
4 Improved Methods for Construction Scheduling 97

4.1 Pseudo-Serial Schedule Generation Scheme . . . . . . . . . . . . . . . . 97
4.2 Water Wave Optimization Applied to RCPSPs . . . . . . . . . . . . . . . 104
4.3 Iterative Swap-based Limited Depth Search . . . . . . . . . . . . . . . . 106
4.4 Evaluation on the Project Scheduling Problem Library (PSPLIB) . . . . . 111
4.5 Uncertainty Extension for the Project Scheduling Problem Library PSPLIB 116
4.6 Data Fusion and Reactive Scheduling . . . . . . . . . . . . . . . . . . . 124
4.7 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
5 State-of-the-Art in Data Acquisition on Construction Sites 125

5.1 Protocol-based Tracking . . . . . . . . . . . . . . . . . . . . . . . . . . 125
5.2 Machine data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
5.3 Laser Scanning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
5.4 Photogrammetry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
5.5 Barcodes and QR-Codes . . . . . . . . . . . . . . . . . . . . . . . . . . 130
5.6 Radio-based tracking . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
5.6.1 GPS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
5.6.2 RFID . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
5.6.3 Ultra-Wideband Radio . . . . . . . . . . . . . . . . . . . . . . . 131
5.7 Visual Tracking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
5.7.1 Range Imaging . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
5.7.2 Video Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
5.8 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
6 Improved Vision-based Progress Monitoring of Excavation Processes 134

6.1 Excavation Tracking using Point Clouds . . . . . . . . . . . . . . . . . . 134
6.1.1 Laser Scanning . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
6.1.2 Photogrammetry . . . . . . . . . . . . . . . . . . . . . . . . . . 135
6.1.3 Processing of Point Clouds . . . . . . . . . . . . . . . . . . . . . 137
6.1.3.1 Clustering . . . . . . . . . . . . . . . . . . . . . . . . 137
6.1.3.2 Cleaning . . . . . . . . . . . . . . . . . . . . . . . . . 138
6.1.3.3 Volume Calculation . . . . . . . . . . . . . . . . . . . 140
6.1.4 Comparison of Laser-scanning and Photogrammetry . . . . . . . 144
6.2 Video-Based Tracking of Excavation Processes . . . . . . . . . . . . . . 145
6.3 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147
7 Data-driven Reactive Scheduling 148

7.1 Concept . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
7.2 Illustrative Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
7.3 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
8 Case Studies 155

8.1 Office Building . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
8.2 Underground Parking Garage . . . . . . . . . . . . . . . . . . . . . . . . 159
8.3 Hospital Building . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
8.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
10 Contents
9 Summary and Outlook 166
10 References 171
A List of Symbols 210
B Benchmark Results 212

B.1 PSPLIB Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 212
B.2 UPSPLIB Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 217
11
1 Introduction
Unexpected delays on construction sites, caused by internal and external influences, create
huge costs every year and account for up to 30 percent of the overall costs of construction
projects (Koehn et al., 1978). Common problems include coordination of different in-
volved companies, weather, traffic, archaeological discoveries, leftover ammunition from
wars, changes in the client’s wishes, scarce resources, and neighborhood repugnance (As-
saf & Al-Hejji, 2006).
"Construction planning is the most crucial, knowledge-intensive, ill-structured,

and challenging phase in the project development cycle due to the compli-
cated, interactive, and dynamic nature of construction processes." - Halpin
& Riggs (1992)
Due to the immense complexity of large scale construction sites, construction sched-
ules are highly sensitive to any changes in the sequence of activities to be performed.
Those detailed schedules often get rendered invalid after a short time and have to be up-
dated, or new ones have to be created, taking newly arisen circumstances into account. In
order to improve this error-prone process, ways to accompany construction using a sim-
ulation model and optimization techniques are investigated in this thesis to support the
maintenance of the plans. Compared to a-priori simulation this approach requires data
to be recorded on-site, to be able to determine the progress of the individual activities.
Eventually this will cause the circumstances in the simulation model to adapt to the actual
state of the project. This thesis aims to describe the development of the required methods
and tools for the case of urban excavation and shoring projects with strong spatial con-
straints. This includes methods to extract the required data from the construction site in
an predominantly automated way, determine the progress of the individual activities and
make predictions on the further progress. When data in the simulation model is updated
it has to be determined whether re-planning would be worth-while. Since the re-planning
process is very time intensive for a human to perform, optimization algorithms were de-
veloped and analyzed in order to automate this process. Furthermore the uncertainty that
is naturally attached to construction processes needs to be taken into account. The con-
cept of the resulting simulation-optimization cycle is illustrated in Figure 1.1.
12 1 Introduction
Figure 1.1: Flow chart of the optimization-simulation cycle. The process starts with an
initial plan that undergoes a simulation and optimization process. During the execution of
the plan, data is acquired about the progress. Unless the plan execution is done, the per-
formance factors can require updating, which causes another simulation and optimization
run, updating the plans respectively.
1.1 Problem Statement and Scope

A construction project consists of a number of activities that all need to be executed for the
project to be finished. Additionally there may be milestones that have to be met to avoid
fines. Each individual activity is described using a number of features. Those include the
time required for completion, required resources, materials, and predecessor constraints.
Resources include machines, workers, and physical space, while materials include con-
crete, bricks, and steel. The duration of an activity can furthermore be uncertain, hence
subject to an underlying probability distribution.
As there are numerous possibilities of delays occurring during the execution of large
construction projects, plans that are created a-priori are often subject to changes and can
rarely be met in detail. This causes substantial costs for all involved parties, as missed
milestones are fined and delayed project completion causes a decrease in payoff for in-
vestors. Therefore it is desirable to notice problems as early as possible and react in an
adequate manner. This requires real-time data acquisition techniques, such as tracking
of machines, or the progress of certain activities. Results include updated performance
factors, activity durations, or progress ratios. Additionally methods to verify the plans’
validity, given the new circumstances, are required. If the plans become invalid and cannot
be updated easily, ways to generate new plans are of strong interest.
The problem of scheduling activities under precedence and resource constraints, in
order to optimize some criteria is called the Resource Constrained Project Scheduling
Problem (RCPSP) (Wiest, 1962). Figure 1.2 shows a graphical notation of such a prob-
lem used throughout the thesis, including resources, materials and geometric spaces. The
rectangular boxes describe the individual activities of the project while the continuous ar-
1.1 Problem Statement and Scope 13
rows connecting the activities represent precedence relationships. For instance activity A
needs to be executed prior to activity B. The round rectangles represent resources, while
the finely dashed arrows imply resource usages. A number next to the arrows indicates
the number of resources of that kind required. If no number is given, this defaults to one.
In order to make representations of simple examples more lean, the colors and hatching
of the activity rectangles also indicate the resource used. In this description we distin-
guish materials, which do not remain available after use, from resources, which can be
used multiple times. Materials are illustrated using white hexagons, connected by dashed
arrows. Similarly to the resources, numbers next to arrows, represent amounts. Finally
the overlapping parallelograms indicate geometric resources. The overlaps indicate which
areas or volumes overlap. A feasible solution to this kind of problem can be illustrated
as a Gantt-chart showing the start and end times of the individual activities (Clark et al.,
1922).
Figure 1.2: Scheme of a Resource Constrained Project Scheduling Problem (RCPSP).

Rectangles represent activities, while their fill color and hatching indicates the resource
that is required. The hexagons represent required materials while the rounded rectan-
gles represent resources. The parallelograms represent geometric spaces. Dotted arrows
indicate usages, while solid line arrows represent precedence constraints.
Additionally uncertainty is an important factor in the planning process of construction

projects. Different soil conditions may cause varying performance of excavators or drills,
while traffic congestion influences the time dump trucks need to transport excavated soil
to a dump site. This kind of uncertainty needs to be included into the simulation and op-
timization process and in this way, can give rise to the robustness of an existing schedule.
This can be done by representing the uncertainty using probability distributions. Further-
more a measure for robustness can be used as part of an optimization objective, forming
a multi-criteria optimization problem. During the construction process, acquired data
can be used to update the associated probability distributions and therefore incrementally
improve the prognosis and allowing to create plans taking even the most recently encoun-
14 1 Introduction
tered circumstances into account.
1.2 Hypotheses
This thesis discusses five main hypotheses:
• Reactive Scheduling can reduce the negative consequences of unexpected delays of

excavation and shoring operations.
• The uncertainty in construction operations can be modeled and handled by the pre-
sented methods.
• It is possible to acquire the required data (semi-)automatically from construction

sites.
• Heuristic optimization approaches can be used to create schedules taking the ac-
quired data into account and minimizing the changes due to further delays.
• The new pseudo-serial schedule generation scheme offers a good basis for meta-
heuristic algorithms to solve RCPSPs.
1.3 Methodological Approach

This thesis focuses on reactive scheduling for excavation and shoring operations, taking
uncertainty into account. In order to acquire data from real-life construction sites, a new
approach, based on photogrammetry was developed and extensively field-tested. Further-
more it has been combined with video-based methods developed in cooperation with the
Georgia Institute of Technology in order to get more detailed performance data on the
individual resources involved in the excavation processes. Additionally an extension of
the resource constrained project scheduling problem has been defined, taking uncertainty
into account. Furthermore a new heuristic optimization approach, capable of obtaining
near optimal solutions for the extended problem, as well as the new pseudo-serial sched-
ule generation scheme (PSSGS), was developed and extensively tested on the commonly
used benchmark problem set PSPLIB (Kolisch & Sprecher, 1997).
1.4 Structure 15
1.4 Structure
This thesis is organized into nine Chapters. The first Chapter defines the problem and
scope, presents the key hypotheses, as well as the methodological approach.
The second Chapter introduces the main elements of construction project models, the
concept of construction simulation, including discrete-event simulation, different repre-
sentations, as well as possible data sources. Additionally it discusses the issue of choosing
the appropriate level of granularity and the associated trade-offs. Eventually the advan-
tages and limitations of the state-of-the-art of construction simulation are discussed.
The third Chapter presents the state-of-the-art in construction schedule optimization

and in that setting defines the resource constrained project scheduling problem (RCPSP),
and describes why it is extended further in order to be applied to real-life scheduling prob-
lems. Then the distinction of active, semi-active, and non-delay schedules is discussed,
followed by existing deterministic and heuristic optimization approaches, followed by the
introduction of uncertainty and robustness into the problem. Then the concept of reason-
able swaps and the Dori algorithm for determining float times are introduced. Eventually
available benchmark problems are presented.
The fourth Chapter introduces four contributions of this thesis. It start with the novel
pseudo-serial schedule generation scheme, followed by its use with the recently published
water wave optimization algorithm by Zheng (2015). Then an improved version of a swap
based local search algorithm, the iterative swap-based limited depth search is introduced.
Eventually a extension of the PSPLIB scheduling problem library is presented, adding
uncertainty to the individual problems. The presented methods are evaluated on com-
monly used benchmark problems. Finally the fusion of the acquired data with reactive
simulation models is discussed.
The fifth Chapter provides an overview of the state-of-the-art in data acquisition on
construction sites. Starting with the traditional protocol-based tracking, it describes the
extraction and evaluation of machine sensor data, as well as different vision-based ap-
proaches.
The sixth Chapter introduces two new procedures for vision-based progress monitor-
ing of excavation sites based on photogrammtery and video analysis. Eventually the two
methods are combined into a coupled system.
The seventh Chapter describes the overall concept of reactive scheduling and how the
presented methods interlock into a syngeric approach. This is illustrated by an example
problem.
The eighth Chapter provides three case studies for the presented algorithms to show
their practical applicability.
The ninth and last Chapter gives a summary and an outlook for future research poten-
tial.
16 1 Introduction
1.5 Terminology
This Section briefly describes important terms used throughout this thesis.
Activity: A process that requires a certain number of resources and specific predecessor
activities to be finished, in order to be executed. It either takes a predefined time to
be finished, or the duration can be uncertain according to certain underlying probability
distributions.
Discrete-Event Simulation (DES): A method for simulating systems, where the state
of the system only changes at discrete points in time, where events happen, such as the
completion of an activity.
Float Time: The amount of time an activity, or a set of activities, can be shifted, or
prolonged within a schedule, without changing the remaining schedule and the overall
makespan.
Makespan: The time difference between start and end time of a schedule.
Material: A non-renewable entity that can be used by an activity when executed, for
instance, concrete, girders, or bricks.
Metaheuristic: An approximate algorithm that aims to efficiently and effectively ex-

plore a search space by the use of a combination of basic heuristic methods (Blum &
Roli, 2003).
Monte-Carlo-Simulation: The act of performing multiple simulation runs of a system

subject to uncertainty in order to obtain statistical measures (Metropolis & Ulam, 1949).
Performance Factor: A measure for the efficiency of a machine, worker, team, or a

combination of those, for instance the amount of soil excavated by an excavator and its
operator per day. Generally this thesis uses the definition of the amount of work performed
over time.
Prerequisite / Predecessor activity: An activity that has to be finished for another ac-
tivity to be executed.
Project: A collection of activities, required and available resources, and constraints that
need to be satisfied for each activities execution.
1.5 Terminology 17
Resource: A renewable entity that can be used by an activity when executed, but af-
terwards becomes available to other activities, for instance, machines, workforce, or geo-
metric areas or volumes.
Schedule: A plan for the execution of a project, assigning start and end times, as well as
resources to each activity. Metrics for further analysis can be extracted from a schedule.
18 2 State-of-the-Art in Modeling and Simulation of Construction Operations
2 State-of-the-Art in Modeling and

Simulation of Construction
Operations
"Simulation is the imitation of the operation of a real-world process or sys-

tem over time. Simulation involves the generation of an artificial history of the
system, and the observation of that artificial history to draw inferences con-
cerning the operating characteristics of the real system that is represented."
- Banks et al. (2001)
The planning process of large construction projects involves large numbers of activ-
ities which are subject to numerous constraints and require different resources and ge-
ometric spaces. The process of analyzing, planning, scheduling, and estimating costs,
taking all factors into account is very tedious and error-prone when performed manually.
Simulation methods are efficient tools to help the involved personnel to reduce errors.
In order to simulate a real system, it has to be represented by an abstract model, that
serves as input for a simulation system. Several methods for modeling of construction
operations have been published in literature, but since computing power has rapidly in-
creased within the last decades, more and more new methods, as well as new data sources
have become available, making construction simulation increasingly interesting for the ar-
chitecture, engineering, and construction (AEC) industry. This Chapter gives an overview
about the evolution of modeling techniques and their feasibility for construction opera-
tions as well as for whole construction projects. As costs and uncertainty play an impor-
tant role in construction planning, methods for representing those factors are discussed.
Another elementary question when modeling a system is the required granularity of the
model and the desired quality of the simulation results. Also this aspect will be taken into
account.
The structure of this Chapter is illustrated in Figure 2.1. On the basis of a real system,
in this case, the construction sites in focus, simulation models are developed in order to
resemble the processes that take place during the execution of the respective construc-
tion projects. The Chapter will introduce process models which describe the nature of
repetitive processes in different ways. These models can be used to acquire metrics about
different aspects of the processes. The different process modeling approaches are de-
scribed in detail in Section 2.5. As an input metric for those process models the efficiency
of the involved resources have to be known. The efficiency of the resources is described
2.1 Performance Factors 19
in terms of performance factors, which are discussed in Section 2.1. The performance
factors form the link between the work to be performed and the time required to do so.
Once the processes involved in a construction project are known, a model of the entire
construction project can be built up using different representations. This process is dis-
cussed in Section 2.6. The process of simulation of the resulting model is described in
Section 2.6.4. The results of the simulation can then be used to improve the execution
plans, as described in detail in Chapters 3 and 4, and in that way feedback into the real
system. Furthermore the performance factors can be dynamically updated using on-site
monitoring systems, as described in Chapters 5 and 6.
Figure 2.1: Flowchart of the reactive modeling and simulation cycle. The real system is
modeled as a collection of process models, which take performance factors of involved
resources into account. The performance factors are estimated and updated by monitoring
the real system. The real system however is subject to uncertainty which in turn affects
the resulting performance factors. The process models are combined into a project model
(See Section 2.6), which is then simulated to generate schedules for the real system.
Once a model has been created it can be used to simulate the system using different
simulation methods. For the case of construction simulation, this Chapter will focus on
methods in the class of discrete-event simulation, which can be used to test different
scenarios or hypotheses, as well as create prognoses for costs and duration of the project.
2.1 Performance Factors

In order to relate work that has to be performed to the time that is required to do so, a
metric for the performance of resources is required. The performance of machines, work-
ers and processes is commonly described by performance factors quantifying the amount
of work performed over time. Pocock et al. (1996) also describes additional performance
indicators such as cost performance, schedule performance, number of contract modifi-
cations and safety information. Atkinson (1999) referred to the factors time, cost, and
quality as The Iron Triangle. For the scope of this thesis, the productivity of machines
and workers is the only type of performance factors in focus, as it unites time and costs.
In case of shoring projects the performance of excavation and shoring operations are vi-
tal (Park et al., 2005). The performance of excavation operations τe can be measured in
volume excavated over time as described in Equation 2.1. Another example where such
performance factors also play an important role is tunnel construction (Grima et al., 2000;
Špačková & Straub, 2013; Yagiz et al., 2009).
∆v vt − vt1
τe = = 2 (2.1)
∆t t2 − t1
Navon (2005) followed a more general approach and analyzed the performance of
construction projects based on four criteria. The positions of earth moving equipment
and personnel was tracked. The movement of tower cranes was recorded and orders and
deliveries of materials were tracked. Additionally the status of guard rails was monitored
to ensure safety of the site. Generally performance was measured as the amount of work
performed in a certain time. A downside of Navon’s approach is that the measurements
were very coarse and no absolute quantitative measurements of earthwork processes were
taken. Navon (2007) presents an approach to obtain more detailed measures of excavation
processes. The model still lacks absolute measurements of the excavated volume, such as
those presented in Chapter 6. Furthermore performance factors are almost always subject
to uncertainty (Baloi & Price, 2003; Salem & Zimmer, 2005; Szczesny et al., 2013).
2.2 Modeling Uncertainty in Construction Operations

In large scale construction projects even small deviations in the individual processes can
cause problems in the synchronization of activities. Besides the performance factors also
durations, idle times, delivery times or the occurrence of certain events, such as contract
negotiations, are subject to uncertainty. Therefore the uncertainty of the individual pro-
cesses must be addressed when modeling such large projects. An early approach for this
problem, named Program Evaluation and Review Technique (PERT), was developed by
the US Navy in order to forecast progress in weapons research and development programs
(Malcolm et al., 1959). It uses a three-point estimation to model beta distributed process
durations. PERT is addressed in detail in Section 3.3.4. Riley (2012) describes the short-
comings of this approach in a practical context and the need for improved approaches.
A more detailed representation of uncertainty can be realized through the generation
of parameters for suitable probability distributions from real-life sample data. Such an
approach was presented by Szczesny et al. (2013) and Szczesny & König (2015). Addi-
tionally the sample data does not have to be based on process durations, but instead can
model the uncertainty in terms of the previously introduced performance factors, model-
ing the uncertainty in the performance of the involved resources. Brandt & Bode (2014)
2.2 Modeling Uncertainty in Construction Operations 21
presented such an approach based on the use of the fuzzy sets possibility theory by Zadeh
(1978) in the work on project scheduling by Masmoudi & Haït (2013).
The following subsections describe the four commonly used probability distributions,
their characteristics, and their suitability for representing uncertain activity durations and
performance factors.
2.2.1 Beta Distribution
MacCrimmon & Ryavec (1964) described that probability distributions used for construc-
tion operations should be continuous and limited between two extremes. Furthermore
AbouRizk & Halpin (1992) showed that common earthmoving operations can be de-
scribed by a beta distribution, which also satisfies the aforementioned criteria. AbouRizk
et al. (1991) studied this matter deeply and developed the Visual Interactive Beta Estima-
tion System (VIBES). VIBES uses a combination of four statistical characteristics of the
process durations to be modeled in order to approximate a beta distribution for processes
where no detailed sample data is available. Two of the input characteristics have to be the
extreme points, the minimum and maximum durations that are possible to happen, while
the other two, have to be two of the following characteristics: mode, mean, variance, or
selected percentiles. VIBES then returns a parameter set for defining a beta distribution
that approximates the provided data. These parameters are the beta distribution’s shape
parameters α and β, and the extreme values. If detailed sample data is available, the
parameters can also be estimated using other methods, such as Maximum Likelihood Es-
timates (MLE) (AbouRizk et al., 1994). The definition of the probability density function
is given in Equation 2.2. The mean of the distribution is defined in terms of α and β in
Equation 2.3.
xα−1 (1 − x)β−1
f (x | α, β) = R 1 (2.2)
0
uα−1 (1 − u)β−1 du
α
µ= (2.3)
α+β
As can be seen in Figure 2.2, manipulating the shape parameters causes the shape
to change in different ways. For all cases an example process with a minimum duration
of 2 minutes and a maximum of 6 minutes was chosen. Increasing the α value causes
the mean to shift towards the maximum possible duration, while increasing the β value
causes the mean to shift towards the minimum possible duration. Fente et al. (1999)
showed that those properties of the beta-distribution are ideal for modeling the uncertain
durations of construction operations, and due to the natural limitations in such processes
both parameters must be positive and larger than one.
2.2.2 Normal Distribution
Golenko-Ginzburg & Gonik (1997) additionally consider normal and uniform random
variables as activity durations. As the normal distribution is not limited between two
α=3, β=3 α=3, β=3

α=4, β=3 α=3, β=4
α=6, β=3 α=3, β=6
α=8, β=3 α=3, β=8
α=10, β=3 α=3, β=10
1 1
0.8 0.8
Likelihood
Likelihood
0.6 0.6
0.4 0.4
0.2 0.2
0 0
0 2 4 6 8 0 2 4 6 8
Duration (minutes) Duration (minutes)
Figure 2.2: Shape change of beta distribution probability density function when manipu-
lating the shape parameters α (left) and β (right).
extremes it needs to be truncated in order to prevent the possibility of too high, too low, or
even negative activity durations. Figure 2.3 shows the effect of the normal distribution’s
parameters mean µ and standard deviation σ on the probability density function given
in Equation 2.4, when being truncated. Same as in the previous example, the Figure
shows activity durations limited between two and six minutes, while shifting the mean or
changing the standard deviation σ. With a sufficiently large value for σ, the results will
get closer to a uniform distribution where each value with in the bounds is as likely as any
other. In order to estimate the parameters σ and µ of the normal distribution, Equations
2.5 and 2.6 can be used.
1 (x−µ)2
f (x | µ, σ) = √ e− 2σ2 (2.4)
σ 2π
N
1 X
µ= xi (2.5)
N i=1
v
u
u1 X N
σ=t (xi − µ)2 (2.6)
N i=1
µ=2, σ=1 µ=4, σ=1

µ=3, σ=1 µ=4, σ=2
µ=4, σ=1 µ=4, σ=3
µ=5, σ=1 µ=4, σ=4
µ=6, σ=1 µ=4, σ=5
0.8 0.5
0.4
0.6
Likelihood
Likelihood
0.3
0.4
0.2
0.2
0.1
0 0
0 2 4 6 8 0 2 4 6 8
Figure 2.3: Change of truncated normal distribution probability density function when
manipulating the parameters µ (left) and σ (right).
2.2.3 Gamma Distribution
Klein & Baris (1991) propose the use the Gamma and Weibull distributions in order to
model service times, which is a very similar use case than construction activity durations.
The exponential and the χ2 distributions are special cases of the gamma distribution and
therefore do not need special treatment. The gamma distribution is defined by the prob-
ability density function defined in Equation 2.7 and uses the gamma function as defined
in Equation 2.8. The gamma distribution is only defined for positive values of x, but
there is no upper bound. Therefore it is truncated in order to represent an upper limit on
the activity duration. The shape parameter k determines the skewness of the distribution,
while the scale parameter θ scales the distribution along the x parameter. This behavior is
illustrated in Figure 2.4. The parameters of the gamma distribution can be estimated by
making use of the fact that the mean µ is defined as the product of the two paramters k
and θ as is shown in Equation 2.9.
x
xk−1 e− θ
(
θk Γ(k)
x ≥ 0 and k, θ > 0,
f (x | k, θ) = (2.7)
0 otherwise
Where:
Z ∞
Γ(t) = xt−1 e−x dx (2.8)
0
µ = kθ (2.9)
k=1, θ=1 k=3, θ=1

k=3, θ=1 k=3, θ=1.25
k=7, θ=1 k=3, θ=1.5
k=9, θ=1 k=3, θ=1.75
k=11, θ=1 k=3, θ=2
1 0.5
0.8 0.4
Likelihood
Likelihood
0.6 0.3
0.4 0.2
0.2 0.1
0 0
0 2 4 6 8 0 2 4 6 8
Figure 2.4: Change of truncated gamma distribution probability density function when
manipulating the parameters k (left) and θ (right).
2.2.4 Weibull Distribution
Similar to the gamma distribution the Weibull distribution (Weibull, 1951) is defined by a
shape parameter k and a scale parameter λ. Both parameters a positive real numbers. A
value for k < 1 indicates that the probability decreases over time, while k = 1 implies
that the probability of remains constant, and k > 1 implies that the probability increases
over time. In terms of construction activities this means, that an activity is very likely
to be finished in minimum time, is likely to be finished within a certain time window, or
is likely to take close to the maximum time, but possibly finish earlier. The mean of the
Weibull distribution is defined in terms of λ and k in Equation 2.11 and makes use the
gamma function defined in Equation 2.8.
(
k x k−1 −(x/λ)k

λ λ
e x ≥ 0,
f (x | λ, k) = (2.10)
0 otherwise
1
µ = λΓ(1 + ) (2.11)
k
k=1, λ=1 k=1, λ=1

k=2, λ=1 k=1, λ=2
k=3, λ=1 k=1, λ=3
k=4, λ=1 k=1, λ=4
k=5, λ=1 k=1, λ=5
2 1.4
1.2
1.5 1
Likelihood
Likelihood
0.8
1
0.6
0.5 0.4
0.2
0 0
0 2 4 6 8 0 2 4 6 8
Figure 2.5: Change of truncated Weibull distribution probability density function when
manipulating the parameters k (left) and λ (right).
2.2.5 Goodness-of-fit Tests
If detailed sample data is available for an activity or a class of activities, goodness-of-fit

tests can be used in order to find the best matching probability distribution. If only limited
data is available, methods such as VIBES can be used to estimate a beta distribution, as has
been proposed by AbouRizk et al. (1991). Most commonly, either the χ2 -Test (Plackett,
1983; Yates, 1934) or the Kolmogorov-Smirnov-Test (Smirnov, 1948) are used in order
to verify whether a distribution fits the given data. For illustration purposes sample data
has been generated for each of the presented probability distributions. Figure 2.6 shows
10, 000 samples of each of the distributions.
Beta Distribution Normal Distribution

1 1
Value
Value
0.5 0.5
0 0
0 5000 10000 0 5000 10000
Sample # Sample #
Gamma Distribution Weibull Distribution

1 1
Value
Value
0.5 0.5
0 0
0 5000 10000 0 5000 10000
Sample # Sample #
Figure 2.6: Sample data of four probability distributions. Beta distribution α=10, β=3
(Top left), Normal distribution µ=0.5, σ=0.1 (Top right), Gamma distribution k=7, θ=1
(Bottom left), Weibull distribution k=5, λ=1 (Bottom right). The plots show 10, 000
samples per distribution.
In order to show which distribution matches the given data, all samples were normal-
ized to the range of 0 to 1, such that all distributions are feasible candidates for all sample
sets. Then the χ2 -Test, as well as the Kolmogorov-Smirnov-Test have been applied to
test every sample set against every candidate distribution. Each test returns whether the
hypothesis, that the data stems from the tested distribution, is accepted or rejected, based
on a p-value, which is the probability that the data, or more extreme (unlikely) versions of
it, are samples of the tested distribution. The parameters of the distributions, the data was
sampled from, are considered unknown and therefore not used for the tests. Both tests
require a parameter α∗ which is the probability of accepting a false hypothesis.
Tables 2.1 and 2.2 give the p-values resulting from the performed tests. Each column
is labeled with the distribution the data is tested for, and each row is labeled with the
XXX
XXXHypothesis
XXX Beta Normal Gamma Weibull
Actual XXX
Beta 0.6617 0.0000 0.0000 0.0000
Normal 0.0000 0.9970 0.0000 0.0000
Gamma 0.0000 0.0000 0.5964 0.0000
Weibull 0.0000 0.0000 0.4979 0.4977
Table 2.1: p-values of χ2 Goodness-of-fit Test.
XXX
XX Hypothesis
XXX Beta Normal Gamma Weibull
Actual XXX
X
Beta 0.9387 0.0000 0.0000 0.0014
Normal 0.3531 0.8613 0.0000 0.0000
Gamma 0.0000 0.0000 0.9720 0.0000
Weibull 0.0061 0.0000 0.9506 0.9632
Table 2.2: p-values of Kolmogorov-Smirnov goodness-of-fit test.
distribution the data has actually been sampled from. As can be seen, the beta, normal
and gamma distributed samples can be easily identified with very high p-values. The
Weibull data on the other hand, could as well be sampled from a special case of the
gamma distribution as a comparison of Figures 2.4 and 2.5 suggests. While, the chi
square test fails to attribute the data to either distribution, the Kolmogorov-Smirnov-Test
returns positive results for both cases, but gives a higher p-value for the true case. All
tests were performed using an α∗ value of 0.05.
This shows how goodness-of-fit tests can be used to identify a well fitting probability
distribution to describe sample data. The resulting probability distribution parameters
form a vital part of a simulation model including uncertainty.
2.2.6 Monte-Carlo-Simulation
Given a problem with an associated solution, based on complex interaction of uncertain

values, it can be difficult to obtain metrics on the solutions that are otherwise likely to
happen. Obtaining a single outcome can yield a non-representative result. Monte-Carlo-
Simulation describes a method that is based on obtaining a large number of outcomes
while sampling new values for the uncertain values for each run (Metropolis & Ulam,
1949). This sampling is performed based on the underlying probablity distributions.
Eventually statistical data is calculated from the collected data. This includes for in-
stance the mean, standard deviation, and worst or best case solutions. Kroese et al. (2014)
describes Monte-Carlo-Simulation as an important tool to gain insight into randomness.
2.2.7 Conclusion
The last Sections provided an overview of the four most commonly used probability dis-
tributions. Literature in the scheduling domain most commonly proposes the use of the
beta distribution for representing uncertainty in process durations. As the beta distri-
bution is bounded, sampled values cannot exceed a minimum and a maximum activity
duration. Furthermore the parameters can be tuned to represent activities that are either
more likely to finish early or late. As the performance factors directly correlate with ac-
tivity durations, this representation is considered equally good for this purpose. Due to
its high suitability, uncertainty is represented using the beta distribution throughout this
thesis. Especially the extension of the Project Scheduling Problem Library PSPLIB, the
UPSPLIB, which is presented in Section 4.5, is built upon the beta distribution. Further-
more the Monte-Carlo-Simulation method forms an important tool for dealing with those
uncertain problems.
2.3 Modeling Costs of Construction Operations

Besides performance, uncertainty, and robustness, the costs are a main criterion for the
success of a project and therefore form another important factor when modeling construc-
tion projects. Almost every activity that is performed on a construction site creates costs.
Those costs can in part be modeled as attributes of the respective activities . On the other
hand there are many fixed costs, such as working crews having to be paid, independent
of their actual work load. However, not all costs are directly associated to construction
activities. There are milestones that have to be met and if missed, cause fines. This can
be solved by introducing milestones as functional elements into the model. When they
are executed prior to their deadline, they create no costs, but afterwards, a fine is added.
Missing a milestone can additionally create running costs, such as a part of a building that
cannot be used yet, and therefore there is no rent being paid.
Furthermore additional resources, such as more machines or workers may be deployed
on the site, in order to speed up the process. Hiring new machines might involve one time
costs for transportation and running costs for rent, fuel and expendable materials. When
a resource is not required for a certain amount of time, it may be worthwhile to free it to
be used on another site, to save costs. For short periods of time on the other hand, the
transportation cost may exceed the savings.
Different kinds of materials may lead to different production times of building ele-
ments, such as different types of concrete being used. In that way, there is a correlation
between time and costs, creating a trade-off problem as described by Li & Love (1997)
and Feng et al. (1997). Adeli & Karim (1997) makes use of this correlation in their neu-
ral dynamics model, by finding a global optimal schedule and subsequently minimizing
the costs for this schedule. Bogenstätter (2000) even included the costs for the entire
life-cycle in the early design phase. All these considerations form a strong motivation to
include costs into the modeling process of construction projects.
2.4 Granularity 29
2.4 Granularity
While it is theoretically possible to implement a continuous simulation model, describing

the precise movement of each machine, personnel, and materials, it is hardly feasible on
the large scale. Therefore the granularity (Level of detail) of the model is an important
decision to be made. This decision forms a trade-off problem between the effort to create
and maintain the model, the effort to work with the model in terms of computational power
and the precision of the results. In order to model construction projects, the granularity is
mainly determined by the aggregation of individual activities.
For instance a bored pile wall can be modeled by the individual activities required to
produce a single bored pile, the production of a single bored pile can be modeled as one
activity requiring all involved resources, or the production of the entire bored pile wall
can be modeled as one aggregated activity. The main downside of aggregating activities
is the coarser representation of the usage of required resources. While the drilling rig is
only required for the activity of drilling the hole, it would be blocked for the entirety of the
bored pile activity when aggregated. Same holds for blocked areas on the construction
site. In practice though, moving resources between different areas on a site, during a
dedicated activity is rarely effective, and not commonly done in practice.
Brooks & Tobias (1996) defined criteria for selecting the best model to represent a
given problem. They define four main criteria: Results, future use of the model, veri-
fication and validation, and required resources. The results should be accurate and ad-
equately describe the behavior of interest. If it is in focus of the user to determine the
optimal order of the bored pile production, then modeling each involved activity individ-
ually can be helpful as it for instance allows exploitation of geometric features to speed
up the progress. When an entire large scale construction project is modeled, the possible
time savings of an optimized bored pile wall production might be negligible, while the
model might have an unfeasible size without aggregation of activities. When there is no
major interaction between the bored pile wall production and other concurrent activities,
the bored pile wall production might be modeled and optimized individually and then be
included in the model of the entire project as one aggregated activity.
As most construction sites are unique, future use of models is rarely of interest. On the
other hand many packages of activities are used on many occasions, and could therefore
be formed into (parametric) modules or process patterns (Sigalov & König, 2015) that can
be reused on other sites. Such a method was presented by Wu et al. (2010) in the domain
of bridge construction and in a more general context by Benevolenskiy et al. (2012). It
would therefore be an option to for instance create a bored pile wall module, that can be
parameterized in order to be used with different depth, thickness, wall length, or drilling
rigs. Furthermore Wu et al. (2010) used the notion of Levels-of-Detail (LoD), that allows
to get an model at a desired degree of abstraction. Verification and validation of the
simulation results can be either be done using historical data, or by tracking excavation
sites to measure how successful the plans, based on the model, can be put into practice.
The required resources are computational power, and time required by the planners
to build the model . A modular approach can simplify the process drastically, but might
require more fine tuning. Creating a coarse model and refining critical activities has been
shown to be an effective approach (Panchal et al., 2008). Schruben & Yücesan (1993) has
shown that the complexity of simulation models can be determined using graph theoretic
approaches. It was shown that several graph theoretic measures can be used on the event
graph to measure the complexity of a model (Brooks & Tobias, 1996). Furthermore, the
characteristics of the resulting model can be formally analyzed as described in Section
3.4.2.
2.5 Modeling and Simulation of Construction Processes

There are copious modeling approaches in literature, each having different notations, dif-
ferent scopes and different advantages. The following Sections introduce different ap-
proaches that have been published throughout the history of computational modeling and
simulation. The capabilities for representing construction projects are clarified using an
illustrative example of a system producing a bored pile wall.
2.5.1 The Bored Pile Wall Example
In order to illustrate the difference between the presented modeling approaches a simple
example of a common procedure in shoring operations, the production of a bored pile wall
has been chosen. A bored pile wall is commonly produced by repeating five sequential
steps. First the drilling rig moves to the position where the next pile should be drilled.
Then the drill is aligned to be exactly horizontal before the drilling process starts. Then
the hole is drilled. After drilling, in most cases a cage of reinforcement steel is lowered
into the hole, which is eventually filled with concrete. Before the hole can be filled with
concrete the drilling rig needs to move away from the hole. For simplification, the re-
inforcement step is not included in this example. The example procedure is depicted in
Figure 2.7. It involves two machines, the drilling rig and the concrete truck. The area
around the hole to be drilled can be blocked by either of the machines. In large con-
struction projects the production of bored pile walls can be parallelized using multiple
machines, implying that the order of the piles to be drilled is not fixed. For overlapping
bored pile walls, a partial drilling order is given. Also drilling next to a freshly drilled pile
is discouraged, as the wet concrete would pour over. For simplification this is not taken
into account in this example.
2.5.2 General Purpose Systems Simulation (GPSS)
IBM described the first general purpose systems simulation (GPSS) for computation pur-
poses in the early 1960s (Gordon, 1961). The simulation models are created using a
graphical modeling language. It consists of 26 different block types that can be used
to generate block diagrams describing the system to be studied. There are entering and
removing blocks, such as the "Call" and "Terminate" blocks. The "Call" block enters
transactions from a list of available transactions into the simulation model at predefined
time intervals or whenever the transaction is accepted by the following block. When a
2.5 Modeling and Simulation of Construction Processes 31
Figure 2.7: Bored pile wall example illustration. The drilling rig is positioned above the
drill location (a), then the rig adjusts to be exactly upright (b), the hole is drilled (c), and
then filled with concrete (d). This illustration was inspired by https://de.wikipedia.org/
wiki/Datei:Foundation_pile_scheme.svg
transactions enters a terminate block it is removed from the simulation. The "Advance"
block represents a simple activity that is executed whenever a transactions enters into it
and takes a certain amount of time. A "Gate" block either accepts or blocks an incoming
transaction depending on the availability of a resource. A loop with a preceding dummy
"Advance" block creates a loop waiting until a resource becomes available. The number
on the right side of a "Gate" block selects the resource to be queried. A "Seize" block
describes the action of seizing a resource until a "Release" block releases it again. The
number in the right side triangle names the resource to be seized or released. A "Hold"
block is similar to an "Advance" block, but during execution holds on to the resource
named in the right side double triangle. Each block in the diagram is uniquely labeled by
a number.
The bored pile wall example is illustrated as a GPSS block diagram in Figure 2.8.
It consists of 15 blocks. The first block, a "Call" block, contains a list of piles to be
produced. Then a "Gate" block causes the system to wait for the drilling rig to be ready,
then seizing the drilling rig in the "Seize" block 3. The "Advance" and "Gate" blocks
4 and 5 wait for the area around the location to be drilled to be free, the "Seize" block
6, then seizes the location, after which the drilling rig moves to the location using the
"Advance" block 7. Block 8 adjusts the drilling rig and block 9 drills the hole. After that
block 10 releases the rig and blocks 11 and 12 wait for the cement truck. Then the "Hold"
block pours concrete into the hole, while holding the cement truck resource. Finally the
location is released in block 14 and the pile is terminated in block 15.
This way of describing the example delivers a good representation of the system. As
soon as the drilling rig is released in block 10, the "Call" block 1 can enter the next pile
into the simulation and it can be drilled while the previous hole is still being filled with
concrete. Hence it allows for parallel execution of the involved processes. Since GPSS is
a general purpose model it has a very high level of abstraction making it difficult to create
valid representations for non-experts. Furthermore it becomes much more complex when
more flexibility is required, such as multiple drilling rigs being available.
Figure 2.8: Bored pile wall example as a block diagram in the notation for the General
Purpose Systems Simulation (GPSS). The call element on the top left outputs all the piles
that need to be produced one by one. following the arrows, the pile is then produced and
is finished once the pile instance reaches node 15, which terminates the production of that
pile.
Figure 2.9 shows the GPSS model extended by an additional drilling rig. For this
purpose 3 more block types are used. The "Branch" block allows to send resources along
either of the outgoing arrows. It allows statistical selection among the possible choices, or
using the first outgoing arrow that is not blocked. The "Tag" block allows to tag a resource
with a number, and the "Transfer" block redirects the resource to an output according to
a previously assigned tag. The updated diagram adds a "Branch" block number 2, that
selects either drilling rig 1 or 2, then seizing the respective rig in blocks 4 or 7. Further-
more the current pile is tagged with the assigned drilling rig. When the drilling rig is to
be released the assigned tag is used in the "Transfer" block 15, to release the respective
drilling rig. As can be seen in this example, adding additional resources to an existing
diagram is troublesome and requires tedious work. Adding multiple cement trucks would
add even further complexity. When multiple resources, with multiple instances are to be
used at the same time, the concept of tagging resources quickly becomes very confusing
and error prone. Understanding such a model as a non-expert can be very difficult. For
Figure 2.9: Bored Pile Wall Example as a block diagram in the notation for the General
Purpose Systems Simulation (GPSS) with two drilling rigs. The call element on top out-
puts each pile that needs to be produced one by one. In node 2 a drilling rig is selected for
each pile, after which the respective rig is used for production, creating the two parallel
streams for seizure of the rig. Once the drilling is seized, the streams merge again, pro-
ducing the pile, and eventually releasing the seized rig. Eventually each pile’s production
terminates in node 22.
this reason more specialized models were developed.
2.5.3 Activity Cycle Diagram (ACD)
The Activity Cycle Diagram (ACD) was introduced by Tocher (1963) to model interac-
tions of system objects. It consists of two states, idle and active, that can be used to de-
scribe simulation entities. different arrows among the entities describe the state changes
of resources. Hence, the arrows indicate the flow of the resources between active and idle
states and therefore indicating the assignment of resources to the individual processes. In
order for a state to become active, all incoming idle states must be holding the required
resources.
Figure 2.10: Bored pile wall example as an Activity Cycle Diagram (ACD). This diagram
shows the flow of the resourced as differently colored arrows. First a hole is selected in
the "next hole" idle state, going into the "DRILL" active state, requiring the drilling rig
to be in the idle state "ready". Once drilling is done, the drilling rig can move to the idle
state "done 1", while the drilled hole can move to the idle state "done 3". Then the drilling
rig enters the "MOVE" active state, freeing the drilled hole, which goes into the "done 4"
state. The drilling rig moves into the "done 2" state, being ready to be adjusted for the
next hole. The drilled hole, can now be filled with concrete, hence enter the "POUR
CONCRETE" active state, given the cement truck is not currently filling another hole,
hence is in the "wait" state. Once the hole is filled, the hole resource can proceed to the
next hole.
For the bored pile wall example, depicted in Figure 2.10, three types of arrows were
introduced, forming three individual cycles for each of the involved resources, the drilling
rig cycle using red solid arrows, the cement truck cycle using green dashed arrows, and
the drill hole cycle using blue dotted arrows. The drilling rig moves along the cycle of
the active states "MOVE", "ADJUST", "DRILL". The cement truck moves along the
cycle "MOVE", "DRILL", and the drill hole cycles between the active states "DRILL",
"MOVE", and "POUR CONCRETE". When the drilling rig is in the idle state "done 1"
following the active state "DRILL", it can commence to move to the next drilling position.
After having moved it continues to the idle state "done 2", after which it directly goes into
the active state "ADJUST". After that it is ready for drilling the next hole. Before the
system can enter the active state "DRILL", the idle state "next hole" needs to hold the
"Drill hole" resource, which is only freed after the concrete has been poured. Once the
"DRILL" state has been finished, both the drilling rig and the drill hole resource are freed
and move to the respective idle states "done 1" and "done 3". After that the active state
"MOVE" can be entered, for the drilling rig to move on to the next location. As the cement
truck must be in the idle state "wait" and the drill hole resource moved to the idle state
"done 4" after the drilling rig moved, the active state "POUR CONCRETE" can now be
entered. Once the concrete has been poured, the drill hole resource moves to the "next
hole" state and the cement truck moves into the "wait" state. At this point the cycle can
start over.
As can be noticed this diagram forces the pouring of the concrete to be finished prior
to the next hole to be drilled, which would practically be possible in parallel to the next
drilling process. This is due to the artificial drill hole resource, which is required to
synchronize the concrete pouring cycle and the drilling rig cycle. This shows that the
ACD notation is only feasible for strictly cyclical problems, such as excavation activities,
but in this case does not allow sufficient synchronization of the concrete and drilling cycle.
2.5.4 CYCLic Operations NEtwork (CYCLONE)
The first modeling approach that has been widely applied to construction processes was
published by Halpin (1977) and is called CYCLONE (CYCLic Operations NEtwork).
Compared to the previously described Activity Cycle Diagram (ACD), CYCLONE intro-
duces several new elements that allow the modeling of more complex systems. Compared
to the normal active state, CYCLONE distinguishes between normal and combi elements,
as well as queue and function elements. Furthermore a formal counter element is intro-
duced for statistical evaluation purposes. A normal state has a single incoming arrow
from an arbitrary other element, that when handing over a resource immediately allows
entering that state. A combi state, on the other hand can have multiple incoming arrows,
that all need to hand over a resource in order for the state to be entered. A queue element
collects incoming resources and supplies them through the outgoing arrows. A function
element spawns new resources to be supplied through outgoing arrows.
The bored pile wall example as a CYCLONE diagram is depicted in Figure 2.11.
Beginning in the state "MOVE", the drill rig can immediately enter the state "ADJUST"
after moving to the appropriate position. After adjusting, it will drill a new hole, which is
supplied through the function element "holes 1". After the drilling process is completed,
the drilling rig can move on to the next state, hence the state "MOVE" becomes active. At
the same time the drilled hole is handed over to the queue element "holes 2". Given that
the concrete truck is being held in the queue element "truck", the "POUR CONCRETE"
combi state can be activated. Once the concrete has been poured, the hole is handed to the
Figure 2.11: Bored pile wall example as a CYCLONE diagram. The colors of the arrows
indicate the resource cycle, same as in Figure 2.10. The function element "holes 1" out-
puts the piles that need to be produced. the combi-state "DRILL" is then executed given
the drilling rig is available. Once the drilling is finished the drilling rig moves away in the
"MOVE" active state and adjusts in the "ADJUST" active state. The hole goes into the
"holes 2" queue element, where the holes queue up, waiting for the cement truck. The ce-
ment truck moves to the "truck" queue element when finished and then becomes available
for the combi-state "POUR CONCRETE". Once a hole is filled with concrete, it moves
to a counter element and proceeds to the "piles done" queue element.
counter element, counting the number of finished piles, then it ends in the queue element
"piles done" and the truck is fed back to the queue element "truck". At the same time
the drilling rig can have drilled new holes and can be filling up the queue element "holes
2". Due to the queuing and function element functionality of CYCLONE, the bored pile
wall example cycle can be adequately modeled in this diagram, while the drilling and
concrete cycles being synchronized through the "holes 2" queuing element. If the drilling
cycle advances faster than the concreting cycle, the "holes 2" queuing element will collect
drilled holes, ready to be filled with concrete, while in the opposite case, the concrete cycle
will idle until a new hole is ready to be filled. CYCLONE is still used frequently due to its
simplicity, while still being capable of representing complex scenarios (Huang & Hsieh,
2012). A system called PROject SImulation Dragados Y Construcciones (PROSIDYC),
which is based on CYCLONE has found its way into real life use, where it was able to
facilitate significant improvements in the project execution (Halpin & Martinez, 1999).
2.5.5 STate and ResOurce Based Simulation of COnstruction ProcEsses (STRO-

BOSCOPE)
A decision, project managers are facing regularly is the selection of resources for different
processes. A machine might be able to perform an operation more efficiently, but at the
same time create more costs, or as a downside block a larger area on the construction
site. For this purpose CYCLONE was extended by a forking element, resulting in the
STate and ResOurce Based Simulation of COnstruction ProcEsses (STROBOSCOPE)
(Martinez & Ioannou, 1994). This extension allows multiple different ways of reaching
a certain goal. The fork element allows the resource flow to leave through any one of
the outgoing arrows. This can mean different orders of executing sub-processes or using
different resources for a process, resulting in different performance factors.
Figure 2.12: Bored pile wall example as a multi-mode STROBOSCOPE process diagram.
Compared to the CYCLONE diagram in Figure 2.11, the diagram was extended by the
fork element, offering a choice between moving to two different piles to be drilled next.
For instance, the next one to the left or to the right.
Figure 2.12 shows a STROBOSCOPE process diagram, where the resource flow of the
drilling rig encounters a fork element after drilling, offering two choices of which way to
go to the next hole. In a larger context the fork element can also be used to switch between
multiple resource flows, assigning different processes to different machines. Though it is
very complex to allow the production of the individual piles in an arbitrary order, as all
those piles would need to enter the system at the same time and many fork elements would
need to model the choices. In the example the choice can be viewed as moving to the next
pile that needs to be drilled, either to the left or to the right. If additional piles or machines
would be introduced, the entire model would need to be updated.
2.5.6 Activity-Based Construction (ABC)
Shi (1999) presented the Activity-based construction (ABC) modeling and simulation
method. Compared to the previously described approaches ACD and CYCLONE, ABC
uses an activity-based representation of the construction processes. The representation of
a construction project is reduced to a collection of activities and resources. Each activ-
ity contains certain attributes, including duration of execution, logical dependencies and
resource demands. Compared to ACD and CYCLONE, there is no distinction between
active and idle states as those arise from the constraints, such as resources and precedence
relationships, being met. A process that requires a resource that is unavailable, is natu-
rally forced to be idle. Resources are indicated by colors and arrows indicate precedence
relationships.
Figure 2.13: Bored pile wall example as an activity-based construction diagram. The
diagram only consists of the four activities "DRILL", "MOVE", "ADJUST" and "POUR
CONCRETE". The arrows indicate precedence constraints. The fill color of the rectangles
indicate the required resource. In order to drill, the rig needs to be adjusted, in order to
move to the next location it needs to have finished drilling, and in order to adjust it needs
to have moved to the next location. Before the concrete can be poured, the hole needs to
be drilled and the rig needs to have moved away from the hole.
The bored pile wall example in this notation is illustrated in Figure 2.13. It shows only
four activities in two colors. Green implies that the activity requires the drilling rig to be
available, while the yellow activity requires the concrete truck to be executed. Again,
starting with the "MOVE" activity, the "ADJUST" activity can start directly afterwards.
Immediately after that the "DRILL" activity can commence. After the "DRILL" activity
is finished, the drilling rig needs to move to the next location, hence the "MOVE" activity
needs to be finished, for the "POUR CONCRETE" activity to have satisfied precedence
constraints. Given the concrete truck is available, the activity can be executed. Similar to
CYCLONE, this approach allows for the bored pile wall example to be modeled, allowing
parallelization of the two cycles, while being synchronized through precedence relation-
2.6 Modeling and Simulation of Construction Projects 39
ships. When the drilling rig moves to the first location to be drilled at, only one of the
two precedence constraints of the "POUR CONCRETE" activity is satisfied, only after
the first hole is drilled, and the drilling rig has moved on, the activity becomes executable.
This example shows that ABC offers a much more intuitive way to model construction
operations, that are much more accessible to non-professionals.
2.6 Modeling and Simulation of Construction Projects

As construction projects in practice can hardly be modeled as a collection of cyclic activi-
ties, a more complex approach to modeling entire projects is required. Building up on the
previously described Activity-Based Construction approach, a non-cyclic variation of this
representation is shown in Figure 2.14. In this model, each activity is only executed once,
then marked as done. Therefore each pile to be drilled creates its own individual chain of
activities. This approach is very similar to the event-graph based simulation modeling ap-
proach by Schruben (1983). The shown example assumes there to be just one drilling rig,
and therefore execution of the individual piles is strictly sequential, hence, no two piles
can be drilled at the same time, as each "MOVE" activity needs the previous "DRILL"
activity to be marked as done, as is indicated by the precedence relationships, shown as
arrows. The "POUR CONCRETE" activities are only dependent on the "MOVE" activity,
hence that the drilling rig has moved on to the next drill location.
Figure 2.14: Bored pile wall example as a non-cyclic sequential process diagram. In
this representation, each rectangle is an activity that will only be executed once, and then
marked as finished. The order of execution needs to respect the precedence relationships
indicated by the arrows and the availability of resources, which are indicated by the fill
color of the activity rectangles. The diagram enforces the order the piles are produced in.
This representation, though, is impractical for general execution, as there could be

more than one drilling rig involved in the execution. Furthermore the bored pile wall
process would usually be included in a larger project.
Figure 2.15: Bored pile wall example as a non-cyclic parallel process diagram. Compared
to the diagram in Figure 2.14, this diagram does not enforce the order of pile production
and allows full parallel execution if multiple drilling rigs or concrete trucks are present.
The representation shown in Figure 2.15 allows full parallelization of the individual
drilling and concrete activities, while being embedded in a larger context. First the site
needs to be levelled in preparation, when this is done, drilling rigs can move to the indi-
vidual location where holes need to be drilled. After adjusting and drilling, the rig needs
to "MOVE ON" in order to free the hole for a concrete truck to be able to execute the
"POUR CONCRETE" activity. The additional "MOVE ON" activity is required as it is
not specified where the drilling rig moves after finishing drilling. When all piles are fin-
ished, the pit, now secured by the bored pile wall can be excavated. This representation
allows full parallelization with an arbitrary number of drilling rigs and concrete trucks,
though it does not take into account that the drilling rig, usually blocks an area blocking
not only a single drilling location. The number of parallel paths in the graph is determined
by the number of piles to be produced, that do not depend on each other.
Figure 2.16 additionally includes a simple geometry representation, which on the one
hand does not require the artificial "MOVE ON" activity and on the other hand allows the
Figure 2.16: Bored pile wall example as a non-cyclic parallel process diagram with ge-
ometry. The diagram in Figure 2.15 extended by a representation of geometric space,
indicated by the blue parallelograms. The spaces may be overlapping as the drilling rig
might block multiple holes while positioned above one hole.
representation of overlapping areas. In order to visualize this, the parallelogram shape

is introduced, representing a specific area or volume on the construction site. The dot-
ted arrows indicate activities blocking that geometric space. Therefore only one activity
blocking a certain geometric space can be active at the same time. Geometric resources
are handled exactly like any other resource, except that blocking one geometry might
block multiple other geometries due to overlap. This allows for detailed modeling of
work spaces, machine sizes, and safety zones.
2.6.1 Activities-on-the-Arrows (AoA) versus Activities-on-the-Nodes (AoN) repre-

sentation
In Figure 2.16, the problem is illustrated as a directed acyclic graph, where each node rep-
resents one activity and edges represent precedence relationships. This notation is called
the Activities-on-the-Nodes (AoN) notation. In contrast to this notation, the Activities-on-
the-Arrows (AoA) notation represents activities by edges and the nodes represent states
(Ponz-Tienda et al., 2015). Figure 2.17 shows an AoA representation of the bored pile
wall example. As the AoN representation is more intuitive to be read and both represen-
tations can be transferred from each other (Yang & Wang, 2010), the AoN representation
is used throughout this thesis.
Figure 2.17: Bored pile wall example as an Activities-on-the-Arrows (AoA) diagram.

Each rectangle represents a state and the arrows indicate activities, causing a state change.
2.6.2 Elements of a Construction Project Model
In order to adequately describe a construction project several elements are required, which
need to be acquainted by different attributes. This Section provides an overview of the
different classes of elements and their capabilities.
2.6.2.1 Activities
Activities are actions performed on and around a construction site, contributing to the
overall progress of the project (Zhang et al., 2005b). Each activity is defined by a set of
parameters. The duration specifies how much time it takes to execute the action. Further-
more an activity may require certain resources, such as machines, or materials, such as
concrete, to allow execution. Additionally it may be required that certain other activities
must be finished or in progress, for the activity to be executed. Most activities create
costs by using resources or materials, so these costs are also an important parameter of
an activity. The costs may also be a function over the duration. In order to determine
robustness of schedules, the uncertainty of each activity’s duration is an important mea-
sure (Artigues et al., 2013; Deb & Gupta, 2006). Therefore a probability distribution for
the duration is part of the required parameters and may influence the cost parameter, as a
prolonged activity might increase running costs, like fuel or wages. More precisely those
costs can be modeled on a resource level, such that the activity itself does not create costs,
but the resources and materials that are used do. A milestone can be viewed as a special
kind of activity that does not have costs, duration, or uses any resources, but is dependent
on some other activities to be finished.
2.6.2.2 Resources
Resources describe renewable entities on the construction site that can be temporarily
used by activities. Resources are commonly available on site in limited quantities and
therefore form one of the strongest constraints in construction scheduling. The usage of
resources may be accompanied by costs, such as fuel or wear and tear. Unless modeled
on an activity-level, the costs for using, hiring, and transporting resources are parameters
of those resources.
2.6.2.3 Materials
Materials are non-renewable entities on site, that can be consumed by activities. Despite
materials having associated costs, the amount of materials used is usually constant on
a construction site, hence it does not change when activities are performed at different
times. Including costs of materials into the project model still might be relevant as it is
possible to calculate the cost flow during the execution phase, as well as there might be
strict budgets for the different construction phases.
2.6.2.4 Geometric Space
Every activity that is performed on a construction site will require some space to per-
formed on, that will be unavailable for other activities during its execution. Those spaces
can be represented either as two dimensional areas on the site, or as three dimensional
volumes inside a building. As geometric conflicts form a strict limit on the feasibility of
construction schedules, it is important to include geometry into the project model.
2.6.2.5 Precedence Constraints
Most of the constraints limiting the execution of construction schedules are logical con-
straints, such as a slab requiring underlying columns in order to be produced, or a wall,
that needs to be plastered prior to being painted. This creates start-finish (SF) prece-
dence constraints, as described in Section 3.3.1. Additionally safety considerations might
create further limitations (Melzner et al., 2012) that can for instance prevent concurrent
execution of certain activities. Furthermore soft-constraints can be included, representing
practical aspects of certain activities, such as dirty activities should be finished before
painting the walls (Beißert et al., 2010). A special kind of constraint is the start-start
(SS) precedence constraint which requires another activity to have started execution, but
it does not necessarily have to be finished (Section 3.3.1).
2.6.3 Computer-Aided Design (CAD) and Building Information Modeling (BIM)

as Data Source
Construction processes can be derived from the product that has to be produced. The
required durations are mostly results of product metrics, such as volume or surface areas.
Those metrics can be extracted from digital representations of the product.
For this purpose Hendrickson et al. (1987) developed a system called CONSTRUC-
TION PLANEX, an expert system for the automatic generation of schedules based on
a database of construction techniques and resource requirements. It formed the first ap-
proach towards the creation of a product model based approach to construction schedul-
ing, taking all the most important factors into account, including precedence relationships,
activity durations, and costs.
Another approach to this problem was published by Navinchandra et al. (1988), in-
troducing a system called GHOST, which started from a very optimistic plan with com-
ponents, connected by few or no precedence constraints, then using a knowledge base to
refine that model by introducing constraints to remove violations. This allows to detect
shortcomings in the available models and develop rules for improving future models.
Winstanley et al. (1993a,b) designed a similar system, called OARPLAN, that linked
activities to components. So instead of assuming that one component is created by one
process, multiple activities can be required for a single component or vice versa. The
information required to do so was extracted from CAD models.
AbouRizk & Mather (2000) presented the idea of simplifying the development of
simulation models for earthmoving operations by an integration with CAD modeling.
Chahrour (2006) generalized this idea and developed an integrated CAD-based simulation
model, that allowed planners to enrich the CAD model by data, required for an integrated
simulation model, during the modeling process. Chahrour & Franz (2006) argue that
this step greatly aids the practical usability of simulation models, as CAD systems are a
commonly known environment among construction planning experts.
Building Information Modeling (BIM) is the process of creating digital representa-
tions of existing buildings, or during the planning phase of new buildings. The models,
besides geometry, should contain physical and functional characteristics of the building.
The concept was first envisioned by Law & Jouaneh (1986) and Björk (1989, 1992).
Popov et al. (2006) presented possibilities to use those digital building models during the
entire process of design, scheduling and construction. Lately this idea gained increasing
popularity in the construction industry (Dossick & Neff, 2010), while even more capabil-
ities of BIMs were discovered, including safety checking (Melzner et al., 2012), progress
monitoring (Bosché et al., 2015; Braun et al., 2015a; Matthews et al., 2015), and produc-
tivity assessment (Poirier et al., 2015).
Tauscher (2011) and Tauscher et al. (2014) took the approach of Chahrour & Franz
(2006) into the BIM age by presenting an approach to automatically extract construc-
tion sequences from building information models. The increase in information available
through the BIM approach allowed to minimize the required human interaction for con-
struction simulation modeling.
Dzeng & Tommelein (2004) developed a case-based system, that tries to identify pre-
viously encountered patterns on the basis of a knowledge-base, and apply solutions that
were executed in the past. The identification is performed on the basis of product mod-
els similar to the approach by Fischer & Aalami (1996)1 . Later the idea was developed
further into a system tailored to work with Building Information Models by Sigalov &
König (2015).
Query languages, such as QL4BIM (Daum & Borrmann, 2014) can use spatial con-
1
Up to this point, this Section was in part inspired by Huhnt & Enge (2006).
straints as a filter criterion applied to building information models. The resulting con-
straints are based on geometric relationships, such as a slab resting on a set of columns,
which need to be produced prior to the slab (Braun et al., 2015a). Those kinds constraints
can be used to create a rough precedence graph of a building to be constructed. Figure
2.18 shows an example of a building model with six stories. The generated precedence
graph is displayed next to the building. Each of the nodes represents one activity and the
connections represent precedence relationships. The graph narrows down on each slab, as
the following columns and other elements can only be constructed after the slab has been
produced.
Figure 2.18: Building information model and its automatically derived precedence graph.
Illustration from Braun et al. (2015a).
Marx & König (2011), König et al. (2012), as well as Braun et al. (2015b) showed that
this concept is equally feasible on a larger scale in real-life BIMs. Additionally building
information models can contain information about building materials, as well as building
techniques. This data gives rise to the involved activities, required resources, materials,
durations, and geometric needs.
2.6.4 Discrete-Event Simulation of Construction Projects
Once a representation of a construction project has been modeled, it can be used to gen-
erate a feasible schedule. Since such a schedule is not necessarily good or even optimal,
optimization procedures such as those described in Chapters 3 and 4 are highly useful. As
construction operations naturally consist of many individual activities that start and finish
at specific points in time, they can be described as a finite number of discrete time in-
stances, where activity states are changing. The simulation of a system, which is defined
by a state that changes only at discrete time instances is called discrete-event simulation
(DES). This notion was formalized by Fishman & Kiviat (1967).
Since discrete-event simulation models exactly that kind of systems it is an appro-
priate tool to use when simulating construction projects. In order to run a discrete-event
simulation system, the state of the system has to be defined for each time step and the
times, when the state changes need to be determined. In case of the construction simula-
tion model used throughout this thesis the state of the system is defined as the states of
each activity and the availability of each involved resource.
The precise definition of an entire construction project in these terms is described in
Section 2.6.2. Schruben (1983) describes simulation modeling in terms of event graphs
and introduced the notation shown in Figure 2.19. It shows two events j and k connected
by a transition taking time t and requiring condition (i). The event graph based represen-
tation is an activities-on-the-arrows (AoA) representation as described in Section 2.6.1.
However it has been shown that an activities-on-the-nodes (AoN) representation is more
intuitive for the user and can be transformed to an AoA notation (Yang & Wang, 2010).
This means that the events j and k correspond to the start and end times of an activity,
the transition time t is the duration of the activity, and that (i) describes precedence and
resource constraints.
Figure 2.19: Notation of simulation modeling with event graphs (Schruben, 1983). Two
states j and k are connected by an activity that requires time t and constraints (i) to be
executed.
In order to perform a simulation run, the resource constraints need to be set, by set-
ting a counter for each resource to the number of available units. The simulation clock
is initialized at time t0 = 0. Then a list of activities with no precedence constraints is
assembled. From this list, an activity A is selected and scheduled to be executed at time
t0 = 0. The required resources are allocated by reducing the respective counters by the
required numbers. A new event is generated at the finish time tf (A) of activity A. The
activity is now considered active. Then the list of executable (precedence and resource
feasible) activities is updated and, if possible, another activity is scheduled at t0 = 0, and
an event is generated for the finish time. If no more executable activities are available,
the simulation clock is advanced to the earliest finish time of all active activities, which
defines the next discrete event. As at this event one or more activities end, the respective
resources are freed and the activity is marked as finished and is no longer active. At this
point a new list of precedence and resource feasible activities is generated, and an activity
is selected for execution. If no more activities are executable the simulation clock is ad-
vanced. This is repeated until all activities have been scheduled. This process is illustrated
in the flowchart shown in Figure 2.20. The node Select feasible activity allows for differ-
ent selection rules that select activities from a set of currently feasible ones. The order in
which the activities are scheduled forms a degree of freedom in scheduling. Based on this
degree of freedom different optimization procedures can be deployed as is described in
Chapters 3 and 4. Beißert et al. (2007) and Beißert (2010) introduced the constraint-based
discrete-event simulation system, where the activity selection is performed randomly and
the results are evaluated using Monte Carlo Methods. Other approaches use different pri-
ority rules, resulting in the parallel schedule generation scheme that is described in detail
in Section 3.4.4.3.
Generally the process of simulation, either as a type of schedule generation scheme, or
as constraint-based discrete-event simulation, forms an important basis for many heuris-
tic algorithms solving resource-constrained project scheduling problems, which are in-
troduced in Section 3.4. Furthermore specific metrics obtained through the simulation
process can be used for certain scheduling procedures, such as the float time calculation
presented in Section 3.4.6.
As described in Chapter 3 in terms of schedule generation schemes, this process is not
necessarily capable to find the optimal schedule as defined in Section 3.4.4.1.
Figure 2.20: Flowchart of the discrete-event simulation process. At time t = 0 the system
is initialized with all resources idling. As long as there are unscheduled activities the main
loop continues. It creates a list of feasible activities, selects a feasible activity, executes
it and allocates the respective resources. Once no more activities are feasible, time is
incremented to the next finishing time, activities are finished and the resources freed.
2.7 Summary 49
2.7 Summary
This Chapter introduced the concept of performance factors, the modeling approaches
to represent uncertainty in terms of different probability distributions, representations of
costs as well as the trade-offs to be made when deciding upon the granularity of process
and project models. The state-of-the-art in cyclic process modeling and non-cyclic project
modeling has been laid out in terms of the required components and the different modeling
approaches. Finally discrete-event-simulation approaches to perform simulations of the
created models have been described.
Simulation is a very capable tool to investigate systems theoretically, based on an
abstract description. While very successful in academic research, construction simulation
has not yet been widely adopted by the architecture, engineering, and construction (AEC)
industry for various reasons, including lack of simplicity, as well as time, costs, and skills
required to set up a usable simulation model (Scherer & Ismail, 2011).
A major limitation is the accessibility of simulation systems to the construction pro-
fessional, who is not necessarily a simulation expert. In the construction context, this
process can be reduced to describing the activities required to complete parts of a project,
or a complete project, their durations, precedence relationships, and resource demands, as
described throughout this Chapter. This makes the simulation domain far more accessi-
ble, as opposed to describing the problem using the General Purpose Systems Simulation
(GPSS), or even programming in general purpose programming languages.
Furthermore advances in the field of building information modeling provide novel data
sources, increasing the usability of simulation systems even further. This approach how-
ever is still subject to different limitations. Due to required abstractions the model may
not adequately reflect the reality on the construction site. It is up to the user to decide upon
the trade-off between granularity and design and simulation effort. Furthermore the in-
put data, such as expected activity durations may be inaccurate and lead to discrepancies.
This can however be addressed by a representation of the uncertainty in the durations,
costs, or underlying performance factors. Overall these limitations and trade-offs need
to be known to the user of construction simulation systems in order to create sufficiently
accurate, but feasible simulation models.
Once a model of a project, its individual processes, as well as a simulation approach
has been chosen, there are different ways to generate schedules for the execution of the
building process. As these schedules are vital for the success of a project it is desired
to generate schedules that are as close as optimal with respect to the desired metrics as
possible. Therefore different optimization approaches have been developed and published
in literature which are presented throughout the next Chapter.
50 3 State-of-the-Art in Construction Schedule Optimization
3 State-of-the-Art in Construction
Schedule Optimization
"The construction professional has to be a ’jack of all trades, and master of

all.’ " - Halpin & Senior (2010)
A reactive simulation system needs to dynamically take changed circumstances into

account and react accordingly. As the main surface to act upon is formed by the order
the activities are executed in, schedule optimization algorithms are of large interest in this
context.
Large scale construction sites commonly consist of thousands of individual inter-
twined activities that are constrained by numerous factors, such as precedence relation-
ships, resource sparsity, and a lack of space. The problem of arranging all those activities
in a way that satisfies all constraints, while additionally minimizing a given objective, is
called project scheduling. In practice a construction project schedule has several impor-
tant aspects that need to be kept in mind. This includes the high volatility of construction
processes that require construction schedules to be robust against delays. Critical pro-
cesses need to be identified, as they need special attention. Furthermore the objectives
for optimizing construction schedules need to be chosen accordingly and must reflect all
important aspects, such as duration, costs, and quality.
The process of project scheduling is mostly performed manually in practice, based
on the experience of the planner. This becomes increasingly complex and error-prone
for larger projects. While the formal project scheduling problem can be solved easily,
its extension, taking resources into account, the resource constrained project scheduling
problem, is NP-Hard in the strong sense (Blazewicz et al., 1983). This means that deter-
ministic algorithms are hardly applicable for large scale problems. This Chapter therefore
mainly focuses on heuristic algorithms.
First, optimization objectives and schedule metrics are introduced, followed by a de-
scription of the incorporation of uncertainty and determination of robustness into the opti-
mization process. Then the handling of multiple optimization criteria is discussed. As an
example for the proposed methods, an illustrative anchored bored pile wall project is in-
troduced, that will be used throughout this Chapter. Then the Project Scheduling Problem
and the Resource Constrained Project Scheduling Problem are introduced. Furthermore
the different types of schedules and schedule generation schemes are discussed.
3.1 The Anchored Bored Pile Wall Project Example 51
3.1 The Anchored Bored Pile Wall Project Example

To illustrate the effects of resource sparsity on construction scheduling a simple example
of an excavation pit is introduced. A rectangular pit has to be excavated, but is surrounded
by soil, threatening to cave in on two sides. In order to prevent that from happening two
bored pile walls will be created. As the pit will be excavated to great depth, the bored pile
walls need to be supported by a layer of anchors. The layout of the site is shown in Figure
3.1.
Figure 3.1: Bored pile wall project example illustration viewed from above. A rectangular
pit is to be excavated. To prevent earth from caving in, two sides of the pit have to be
secured using anchored bored pile walls.
It shows the two bored pile walls, which meet in the upper left of the illustration.
Every second pile will be anchored into the surrounding soil to prevent the piles from
falling over. The illustration shows four distinct excavation areas labeled 1 through 4. In
order for the anchors to be placed, the respective bored pile wall needs to be finished, and
the excavation area has to be partly excavated to the depth the anchors should be placed
and in order to make room for the required machinery. After the anchors have been
placed the remaining excavation of the area can proceed. After all excavation activities
are finished the foundation can be produced. The degrees of freedom of this example
consist of the order the excavation areas are excavated and when the bored pile walls are
produced. Areas 1 and 2 depend on the bored pile walls, while 3 and 4 are independent.
3.2 Objectives of Optimization and Schedule Metrics

The most commonly used objective of the optimization of schedules is the makespan, the
overall duration of the schedule. In construction projects on the other hand, the costs are
often more relevant, but also more complex to determine. There are milestones to be met
which are subject to fines when delayed. Furthermore machines create costs, even when
not used, but the transportation of machines from and to the site can also require extensive
efforts. Therefore it can be more economic to have a machine idle for a certain time, as
opposed to moving it to another site for the time it is not required. The costs for the
availability of machines, but also the stocking of materials should be part of cost oriented
optimization.
Another important aspect is the robustness of a schedule, which describes the limited
impact of the changes in durations of individual activities on the overall makespan. Addi-
tionally leveling the usage towards constant utilization of certain resources can be desired,
forming the Resource Leveling Problem (RLP) (Ponz-Tienda et al., 2013).
Generally, optimization problems in the domain of construction scheduling are in
most cases multi-objective problems, such that for instance resource usage and the overall
makespan are objectives to be minimized.
3.2.1 Uncertainty and Robustness
Sabuncuoglu & Goren (2009) and König (2011b) describe proactive and reactive ap-
proaches to dealing with uncertainty of construction operations. Reactive approaches
imply that the schedule is updated when deviations occur, while proactive approaches try
to anticipate the deviations by modeling uncertainty as described in Section 2.2 and there-
fore maximize the robustness. Since construction processes are highly volatile, robustness
may well be among the most critical measures in construction schedule optimization.
König (2011a) defines a robustness of a construction project schedule as minimiz-
ing the effects of forseeable variations of the durations of construction activities on the
schedule. Herroelen & Leus (2004) define it as incorporating a degree of anticipation of
variability during project execution. They furthermore distinguish quality robustness and
solution robustness. Quality robustness means that deviations in activity durations have
minimal impact on the quality of the schedule, which implies that the changes to the value
of the objective, such as the makespan, are minimal. Solution robustness means that the
changes to the schedule that may be caused by deviations are minimal.
Furthermore Chtourou & Haouari (2008) defines twelve measures of robustness of a
given schedule in terms of the slack si , the number of direct successors Nsucci , the number
of resources of type k used rik , the average percentage increase in activity duration κi and
the slack indicator αi defined in Equation 3.1. All measures are listed in Table 3.1. The
first measure M1 maximizes the sum of all float times, implying that a schedule with
more float time is more desirable. If for instance a single activity with high uncertainty
would have zero float time, this would be considered equally good than an activity with
very low or no uncertainty having zero float time. Also activities with a large number
of follow up activities are more desirable to have larger float times, which is expressed
by M2 . M3 instead takes the resource usages into account by making high float times
for activities with high resource usage more desirable, as resources might be blocked by
previous delayed activities and therefore would cause more potential instability. M4 is
a combination of both previous measures. M5 maximizes the number of non-critical
activities. A lower number of activities with zero float time is preferred. M6 again takes
the number of successors into account, while M7 takes resources into account and M8
combines both aspects. M9 minimizes the total average increase in duration, given it
3.2 Objectives of Optimization and Schedule Metrics 53
is lower than the float time for each activity. M10 , M11 , and M12 expand this idea by
taking successors and resources into account.
(
1 if si > 0
αi = ∀i ∈ J (3.1)
0 otherwise
X X X
M1 = si M7 = (αi rik )
i∈J
X Xi∈J k∈R X
M2 = si Nsucci M8 = (αi Nsucci rik )
i∈J
X X Xi∈J k∈R
M3 = (si rik ) M9 = min(si , κi ti )
i∈J
X k∈R X i∈J
X
M4 = (si Nsucci rik ) M10 = min(si , κi ti ) · Nsucci
i∈J
X k∈R i∈J
X X
M5 = αi M11 = (min(si , κi ti ) rik )
i∈J
X i∈J
X k∈R X
M6 = αi Nsucci M12 = (min(si , κi ti )Nsucci rik )
i∈J i∈J k∈R
Table 3.1: The 12 measures of robustness by Chtourou & Haouari (2008)
Another way of determining the robustness of the resulting schedule, if the probability
distributions of the activities’ durations are known or can be determined as described in
Section 2.2, is Monte-Carlo simulation (Wübbeler et al., 2008). This approach is based
on performing multiple simulation runs or schedule generations, while for each run, sam-
pling new duration values from the respective distributions. When sufficiently many runs
are performed, a distribution of the overall makespan forms, which gives rise to the prob-
ability for delays. For instance the probability of certain milestones being met can be
calculated on this basis.
In a more general context Chassein & Goerigk (2016) describe the problem of robust
optimization as a bicriterial problem in which the two criteria are formed by the average
and the worst-case performance. Both of these criteria should be subject to optimization.
Therefore multi-criteria optimization is an important part when aiming to take robustness
into account.
3.2.2 Multi-Criteria Optimization
Creating a schedule with a maximal robustness can be a good objective in construction

schedule optimization, but as a mere objective it would yield schedules which maximize
slack and idle times. This would tend to create unnecessarily long and expensive sched-
ules. On the opposite, a schedule minimizing costs or makespan will tend to be overly
fragile. Furthermore the usage of additional resources can reduce the makespan, while
increasing the costs, implying a trade-off between costs and makespan. It is therefore
important to take multiple objectives into account. As most optimization algorithms are
designed to minimize a single objective function, ways of evaluating the superiority of a

certain schedule, or mapping multiple objectives to a single objective function are desired.
Usually the user has to decide upon which objectives are of more or less importance.A
survey of such methods was published by Marler & Arora (2004), on which the following
Sections are loosely based.
3.2.2.1 Pareto Optimality
In the field of multi-critertia optimization the work of Pareto (1906) plays an important
role, as he defined the terms Pareto Optimality and weak Pareto Optimality. Lemmas 1
and 2 define the terms respectively.
Lemma 1. A point, x∗ ∈ X, is Pareto optimal iff there does not exist another point,
x ∈ X, such that F (x) ≤ F (x∗ ), and Fi (x) < Fi (x∗ ) for at least one function (Marler
& Arora, 2004).
Pareto optimal points are on the boundary of the space of feasible solutions. A point
is Pareto optimal if and only if there is no point having the same or a lower value for all
objectives F , but a lower value for at least one objective Fi . The set of points satisfying
that condition is called the Pareto Frontier. An example of such a frontier is illustrated
in Figure 3.2. The blue crosses are 50 random points and the red line shows the Pareto
frontier.
0.8
0.6
0.4
0.2
0
0 0.2 0.4 0.6 0.8 1
Figure 3.2: Plot of 50 random points and their Pareto frontier illustrated by the thick red
line.
.
Lemma 2. A point, x∗ ∈ X, is weakly Pareto optimal iff there does not exist another
point, x ∈ X, such that F (x) < F (x∗ ) (Marler & Arora, 2004).
A point is weakly Pareto optimal if and only if there is no other point that improves
on all objectives. All points that are Pareto optimal are also weakly Pareto optimal.
3.2 Objectives of Optimization and Schedule Metrics 55
3.2.2.2 Compromise Solution
An alternative approach to Pareto Optimality is the concept of a compromise solution (Yu

& Leitmann, 1974). For this purpose an utopia point, which is a point consisting of the
minimal value for each objective, is created. This is formalized in Lemma 3. Mostly the
utopia point will never be reached, but a point as close as possible to the utopia point
is called the compromise solution (Marler & Arora, 2004). An example of 10 random
points with their utopia and compromise points is shown in Figure 3.3. As closeness
is commonly calculated using the Euclidean norm, the choice of the closest point may
depend on the scale of the individual objective functions. It may be difficult to adequately
scale the objective functions in practice, if it is uncertain how different units, such as time
and costs, are related.
Lemma 3. A point F o is an utopia point iff Fio = min Fi (x) ∀i (Yu & Leitmann, 1974).
x
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0 0.2 0.4 0.6 0.8 1
Figure 3.3: Plot of 10 random points with the light blue star at (0.2; 0.5) indicating the
compromise point. The thin blue lines indicate the minima for each objective, the red star
in the lower left is the utopia point.
.
3.2.2.3 Weighted Global Criterion
For the weighted global criterion all objective functions are combined into a single func-
tion by a weighted exponential sum. The simplest form is given in Equation 3.2. It
involves a vector of weights w which is set a-priori in order to define the importance of
each objective function. The sum of all weights should equate to 1. Furthermore the pa-
rameter p is set proportional to the importance of minimizing the function with the largest
difference to the utopia value Fio (Marler & Arora, 2004). An extension of this method
was presented by Yu & Leitmann (1974) and is described in Equation 3.3. It does not
require the objective function values to be strictly positive, as it makes use of the utopia
point F o in order to offset each objective function.
k
X
U1 = wi [Fi (x)]p , Fi (x) > 0 ∀i (3.2)
i=1
k
X p1
U2 = wip [Fi (x) − Fio ]p (3.3)
i=1
3.2.2.4 Weighted Sum
The simplest and most common approach to multi-objective optimization is the weighted
sum of the individual objectives as given in Equation 3.4. Each objective is assigned a
weight a-priori in order to set its importance. This approach is used for the objectives of
the uncertainty extension of the PSPLIB problem library described in Section 4.5.
k
X
U3 = wi Fi (x) (3.4)
i=1
3.3 Project Scheduling Problem

In the Project Scheduling Problem, a set of precedence-constrained activities has to be
arranged in a way - minimizing a given objective (Möhring & Schulz, 2003). A problem
is defined as an acyclic directed graph, where each node represents an activity and edges
represent precedence relationships. Each activity additionally takes a certain amount of
time to be executed. A feasible solution to a project scheduling problem assigns start and
end times to each of the activities, such that none of the precedence constraints is vio-
lated. It is desired to find a feasible solution that minimizes or maximizes some objective
function, such as the makespan or the project costs.
Compared to the previously discussed approaches, the Project Scheduling Problem
deals with non-cyclic activities, not taking any resources into account. While the Project
Scheduling Problem is not an optimization problem it is the basis of the resource-con-
strained project scheduling problem introduced in Section 3.4 and many optimization pro-
cedures are based on the solution of the associated Project Scheduling Problem. Therefore
the two solution procedures for the Project Scheduling Problem are introduced within this
Section.
3.3.1 Precedence Constraints
In order to define precedence relationships among the activities different kinds of con-
straints can be used. While in Section 2.6.2.5 only one type of precedence relationship
3.3 Project Scheduling Problem 57
is described, there are multiple types of such constraints. Figure 3.4 shows the four ba-
sic types of precedence relationships with an illustrative feasible schedule, showing the
earliest possible start time for B, relative to A, for each one on the right side. The most
common and most intuitive precedence constraint is the Finish-Start (FS) constraint, re-
quiring the first activity to be finished for the next one to start. If not specified otherwise,
this type of precedence relationship is the default one used throughout this thesis. The
second type is the Finish-Finish (FF) relationship, requiring the second activity to finish
at earliest, simultaneously with the first activity. The Start-Finish (SF) constraint requires
the second activity to finish at earliest when the first activity starts. Finally the Start-Start
relationship requires the second activity to start at earliest when the first activity starts.
Figure 3.4: Types of precedence relationships. a) Finish-Start (FS), b) Finish-Finish (FF),

c) Start-Finish (SF), d) Start-Start (SS).
3.3.2 Graphical Representation of the Bored Pile Wall Project Example
A project scheduling problem is defined by an acyclic directed graph of activities as de-

scribed in Section 2.6, but without taking resources into account. Each node represents
an activity and the directed edges indicate the precedence relationships. The precedence
relationship graph for the example introduced in Section 3.1 is shown in Figure 3.5. Each
of the involved bored pile walls creates one strand of four activities. Each strand starts by
producing the piles, which is indicated by the nodes B1 and B2. After the piles have been
produced the first layers of excavation can be performed, represented by the nodes E11
and E12. After that, the walls can be anchored by the activities A1 and A2, allowing the
activities E21 and E22 to finish the excavation. Eventually, after the areas E3 and E4 have
been excavated, the foundation can be produced in activity FND. The box on the lower
right scales the durations of the individual activities which are indicated by the width of
the individual boxes. Activity B1 has a duration of 4 time units, E11 and E12 have a
duration of 5 time units, all other activities have a duration of 6 time units.
Figure 3.5: Bored pile wall project example precedence constraints. The two bored pile
walls (B1 and B2) need to be finished before the first layer of excavation (E11 and E12),
after which the pile walls can be anchored (A1 and A2) followed by the second layer
of excavation (E21 and E22). Furthermore excavation activities E3 and E4 need to be
performed. When all excavation is finished the foundation (F) can be produced.
The shortest possible schedule for this problem is shown in Figure 3.6 as a Gantt dia-
gram. A Gantt diagram is a graphical representation of a schedule of activities over time,
as defined by Gantt (1903). The horizontal axis describes the time while the vertical axis
aligns the activities. Each line corresponds to exactly one activity. The arrows between
the activities illustrate the satisfied precedence constraints. It is visible that both strands
are executed in parallel as the precedence constraints in the example do not prevent that.
3.3.3 Critical Path Method (CPM)
In order to generate solutions for project scheduling problems Kelley & Walker (1961)
described an approach, the Critical Path Method (CPM), which, while in itself not being
an optimization procedure, formed a basis for many other algorithms. The critical path
method is a two-pass approach consisting of a forward and a backward pass. In each
pass start and finish times for each activity are calculated. During the forward pass the
algorithm starts by selecting all start activities, which have no predecessors. In the case
of the example shown in Figure 3.5 these are B1,B2,E3, and E4. These activities are
Figure 3.6: Bored pile wall project example solution.
assigned an early start time tES of 0, and the early finish time tEF is set to the activity’s
duration added to its early start time. Those times are written on the upper left and upper
right of the activity’s boxes respectively. After that, all activities of which all predecessors
have been treated are selected and the maximum early finish time of all predecessors is
assigned as the early start time of each activity. The early finish time is assigned by adding
the duration. Figure 3.7 shows the example after the forward pass has been executed.
Figure 3.7: Bored pile wall project Example after the forward pass of the critical path
method.
After the forward pass is completed, each activity has an early start and an early finish
time assigned. The second stage of the algorithm consists of the backward pass. For this
pass all precedence relationships are considered in reverse order. The latest of all early
finish times is considered to be project horizon. Therefore the latest activities, that have
no successors are selected initially and are assigned the latest early finish time as their
late finish times tLF . Their late start times tLS are calculated by subtracting the respective
duration from the late finish time. After that, all activities of which all successors have
been treated are selected and their late finish times are assigned to be the earliest late start
time of all successors. Their late start times are again calculated using the duration of the
activities. The late start and late finish times are written on the lower right and lower left
of the activities’ boxes respectively. This results in the state illustrated in Figure 3.8.
Figure 3.8: Bored pile wall project example after the backward pass of the critical path
method.
Eventually the total float time, or the slack, si can be calculated for each activity i.
This is done by subtracting the early start time and the duration from the late finish time.
The resulting value represents the amount of time the activity can be shifted or delayed
without changing the overall project duration. Table 3.2 shows the results of the algorithm
for the example. The bottom row shows the total float times for each activity. Each activity
that has a total float time of 0 is considered critical. This means that any delay in those
activities propagates through the entire schedule. The set of all activities with zero total
float time forms one or more critical paths. In this example the critical path is B2 → E12
→ A2 → E22 → F.
The resulting schedule is visualized in Figure 3.9. Each activity is either blue, which
indicates it is critical, or has a gray bar attached, indicating the total float time or the range
B1 E11 A1 E21 B2 E12 A2 E22 E3 E4 F

tES 0 4 9 15 0 6 11 17 0 0 23
tEF 4 9 15 21 6 11 17 23 6 6 29
tLS 2 6 11 17 0 6 11 17 17 17 23
tLF 6 11 17 23 6 11 17 23 23 23 29
si 2 2 2 2 0 0 0 0 17 17 0
Table 3.2: Results of the critical path method for the example in Figure 3.5.
the activity can be shifted in. As the excavation activities E3 and E4 are only required for
the final foundation activity F and do not depend on any other activities to be finished
they can be executed anytime before the foundation activity. Also, since the first bored
pile wall B1 is built much faster than the second one, as it contains fewer piles, it, and all
the follow up activities E11, A1, and E21, that are preceding F, have non-zero float time,
and hence are not critical.
Figure 3.9: Visualization of critical path and total float times for the example in Figure
3.5. The gray bars indicate total float times, while the blue activities are critical.
The critical path method is an exact procedure that generates a feasible schedule for
project scheduling problems with precedence constraints. It does however not consider re-
source sparsity, which is a commonly encountered constraint in real-life scheduling prob-
lems. While the project scheduling problem can be exactly solved using the described
methods, it does neglect the need for resources. Most construction activities require ma-
chines, workers, or geometric spaces. These demands add strong constraints in practical
scenarios, which need to be taken into account. Dori (2016) addressed the determina-
tion of float times and the critical path, taking resource constraints into account, which is
discussed in detail in Section 3.4.6.
3.3.4 Program Evaluation and Review Technique (PERT)
The Program evaluation and review technique (PERT) was developed by the US Navy
in order to forecast the research and development progress in large projects involving a
simple representation of uncertainty (Malcolm et al., 1959). Compared to the Critical
Path Method, PERT aims to find a schedule taking the expected duration of activities into
account. The method is based on modeling each involved activity by a set of parameters.
Those parameters include three time metrics which model the uncertainty of the activ-
ity durations, the optimistic duration tO , the most-likely duration tM , and the pessimistic
duration tP . From these three metrics, a fourth metric, the expected duration tE is cal-
culated using Equation 3.5, which is based on assuming that activity durations are beta
distributed as defined in Section 2.2.1 and tO and tP are the limits of the feasible range of
the distribution.
tO + 4tM + tP
tE = (3.5)
6
Furthermore each activity can have precedence constraints as defined in Section 3.3.1.
All input data is usually represented in tabular form. Such a representation, listing prece-
dence constraints and the four duration metrics for each activity is shown in Table 3.3.
Activity Predecessors tO tM tP tE
B1 3 4 6 4.17
B2 5 6 8 6.17
E11 B1 4 5 6 5
E12 B2 4 5 6 5
A1 E11 5 6 9 6.33
A2 E12 5 6 9 6.33
E21 A1 4 6 9 6.17
E22 A2 4 6 9 6.17
E3 4 6 9 6.17
E4 4 6 9 6.17
F E21,E22,E3,E4 5 6 7 6
Table 3.3: Tabular representation of the example in Figure 3.5 with added uncertainty.
From the data shown in the table, five time metrics are calculated, the early start time
tES , the late start time tLS , the early finish time tEF , the late finish time tLF , and the
slack time tS . The early finish time tEF of an activity is calculated by adding the expected
duration tE to its early start time tES . The early start time is the latest early finish time tEF
of all its predecessors. The latest finish time tLF of an activity is the smallest of all late
start times tLS of all its successors. Activities without successors have a late finish time
tLF equal to the largest early finish tEF time of all activities. The late start times tLS of the
activities are calculated by subtracting the expected duration tE from the late finish time
tLF . The slack time tS is calculated by subtracting the early start time tES from the late
start time tLS of each activity. The calculation of the early start times tES can be viewed
as calculating the longest predecessor path towards this activity, while the calculation of
the late finish time tLF can be viewed as calculating the longest path along the successors
starting from the last activity. An activity with a slack time of zero is considered critical.
Activity tES tEF tLS tLF tS

B1 0 4.17 2 6.17 2
B2 0 6.17 0 6.17 0
E11 4.17 9.17 6.17 11.17 2
E12 6.17 11.17 6.17 11.17 0
A1 9.17 15.5 11 17.5 2
A2 11.17 17.5 11.17 17.5 0
E21 15.5 21.67 17.5 23.67 2
E22 17.5 23.67 17.5 23.67 0
E3 0 6.17 17.5 23.67 11.33
E4 0 6.17 17.5 23.67 11.33
F 23.67 29.67 23.67 29.67 0
Table 3.4: Tabular representation of the results of applying PERT on the data from Table
3.3.
The results are either displayed in tabular form, as shown in Table 3.4, or in a so-
called PERT chart, which can be represented as an Activities-on-the-Node (AoN) or an
Activities-on-the-Arrows (AoA) network. For the AoN representation, nodes are com-
monly illustrated as is shown in Figure 3.10. An AoN representation of the anchored
bored pile wall project example is shown in Figure 3.11.
Figure 3.10: Representation of an activity in the PERT chart notation.
It becomes visible that the path from B2 towards F is critical, while all other activities
have non zero slack times. PERT is an early approach to managing large projects and, by
formalizing the defined metrics, as well as the PERT chart, offered new tools to project
managers. Furthermore through the incorporation of uncertainty, it allowed assessment
of the risk of delays, and therefore allowed to minimize costs. The calculation of the
slack time also helped to identify critical activities. Resource sparsity however, was in the
context of large scale military projects, not in the focus of the developers.
Figure 3.11: Anchored bored pile wall project example solution in PERT notation. The
critical path, consisting of all activities with zero float time (B2 → E12 → A2 → E22 →
F), is highlighted in blue. The corresponding Gantt Chart is shown in Figure 3.9.
3.4 Resource Constrained Project Scheduling Problem 65
3.4 Resource Constrained Project Scheduling Problem

Extending the project scheduling problem by resource demands, creates the Resource-
Constrained Project Scheduling Problem (RCPSP). The object of the RCPSP is to sched-
ule a given set of activities in an order that fulfills all given precedence and resource
constraints while minimizing the overall makespan of the project. A formal definition of
the problem is given in Section 3.4.7. The precedence constraints, same as in the project
scheduling problem, may be represented as an directed acyclic graph, that defines a partial
order of the activities. Each activity additionally may use a number of resources of differ-
ent types. There is a limited number of instances of each resource type. As this problem
is known to be NP-hard in the strong sense (Blazewicz et al., 1983), several heuristic al-
gorithms have been developed, which are presented in Section 3.4.8. However, RCPSPs
are still subject to many ongoing research endeavors.
3.4.1 Graphical Representation
There are different ways of representing the entirety of a resource constrained project
scheduling problem. Due to the different possible cases, different notations are found
throughout literature. When the anchored bored pile wall example from Figure 3.1 is
extended by resources, a possible outcome is shown in Figure 3.12 (Davis, 1973).
Figure 3.12: Bored pile wall project example with precedence and resource constraints.
On the upper right of each of the activities’ boxes there is a vector of the number of
resources of each type that is required for the activity to be executed. For simplicity only
one type of resource, a worker, is considered. For the bored pile wall activities B1 and
B2, three workers are required. Each bored pile is created by drilling into the ground
by connecting parts of a metal casing to each other, to go deeper into the ground. One
worker is required for operating the drilling rig, one for operating the crane, delivering the
casing parts, and one to guide the casing to connect to the previous part. The excavation
activities only require one worker for operating the excavator. The anchoring process
requires two workers for operating and guiding the anchoring machine. The foundation
process requires three workers, adding reinforcement material and concrete.
Another, more intuitive way of representing resources graphically, is the notation
shown in Figure 3.13. In this notation resources are shown as rounded rectangles and
dashed arrows towards these rectangles indicate resource requirements. The numbers on
the arrows indicate the number of instances required. For the anchored bored pile wall
project example, the worker resource is added on the top side of the figure and each ac-
tivity indicates its requirements by a dashed arrow.
Figure 3.13: Bored pile wall project example precedence and resource constraints.
Additionally the activities’ boxes can be colored to indicate the required resources,
furthermore geometry and their overlaps can be indicated by a rhombus shape, and a
hexagon indicates material demands. A general representation of such a problem is illus-
trated in Figure 3.14.
3.4.2 Problem Characteristics
In order to determine how hard a RCPSP is, prior to solving it, it is useful to analyze
it. This Section presents commonly used metrics for characterizing RCPSPs. A number
of indicators characterizing the structure of scheduling problems have been published in
literature. Patterson (1976) defined a number of such indicators and analyzed their impact
on the performance of (meta-)heuristic algorithms. The most commonly used indicator
among which is the resource constrainedness. At the same time Cooper (1976) defined the
resource factor, as well as the resource strength. The resource constrainedness is defined
as the ratio γk of the average of the quantities rik of resource k required by activity i,
Figure 3.14: Scheme of a Resource Constrained Project Scheduling Problem (RCPSP).
requiring at least one unit, as defined in Equations 3.6 and 3.7, and the total number of
units available of that resource rk , resulting in Equation 3.8 (Patterson, 1976).
(
1 if rik > 0
ρik = ∀k ∈ R, ∀i ∈ J (3.6)
0 otherwise
X
rik
i∈J
φk = X ∀k ∈ R (3.7)
ρik
i∈J
φk
γk = ∀k ∈ R (3.8)
rk
The resource factor λi , as defined in Equation 3.9 by Cooper (1976), is the ratio of the
average number of resources required for each activity and the total number of resource
types |R|.
XX
ρik
i∈J k∈R
λi = (3.9)
N · |R|
The ratio of the number of resources available of type k, rk and the average number
required by all activities Φk , as defined in Equation 3.10 is called the resource strength
ξk . It is calculated as is shown in Equation 3.11 (Cooper, 1976).
X
rik
i∈J
Φk = ∀k ∈ R (3.10)
N
rk rk · N
ξk = =X ∀k ∈ R (3.11)
Φk rik
i∈J
For a more detailed overview of problem characteristics the reader is referred to the
papers by Patterson (1976) and Cooper (1976).
3.4.3 The Anchored Bored Pile Wall Project Example
It is now assumed that only 4 workers are available, which for the schedule illustrated in
Figure 3.6 results in the resource profile shown in Figure 3.15. The vertical axis indicates
the number of workers required, while the horizontal axis represents the time. The red line
marks the resource limit of four workers. Such a resource profile is created by stacking all
activities that require the analyzed resource on top of each other. In this case the activities
E11, A1, and E21 are split into two boxes, to prevent gaps in the profile. It is visible that
this resource limit is exceeded by the parallel execution of the bored pile wall activities
B1 and B2 on the very left.
Figure 3.15: Bored pile wall project example resource usage.
Therefore, in order to not exceed the four worker resource limit, the activities B1 and
B2 cannot be executed in parallel. A schedule preventing that is illustrated in Figure 3.16.
Activity B1 is executed first, followed by activity B2. As illustrated in Figure 3.17, the
resource limit is not exceeded at any time throughout the schedule.
However, if instead of B1, the other bored pile wall B2 would be produced first, this
would result in the schedule shown in Figure 3.18, which also does not exceed the re-
source limit, as shown in Figure 3.19. The new schedule is more effective than the previ-
ous one, shown in Figure 3.16, as the duration is shorter. Also it is visible in the resource
profile, that the idle times of workers are reduced. This illustrates how choices in the or-
ders of activities result in different schedules with different makespans. Finding the order
of activities minimizing some objective is in the focus of this Section.
Figure 3.16: Bored pile wall project example solution with resource constraints.

Figure 3.18: Bored pile wall project example solution with resource constraints.

3.4.4 Schedule Generation Schemes
A schedule generation scheme is a formal way of generating a feasible schedule for a

given RCPSP. As many optimization procedures operate on the priorities of the indi-
vidual activities, schedule generation schemes are commonly used to generate schedules
from those priorities in a deterministic way. There are different schemes creating sched-
ules with different properties. The most commonly used schemes are the serial schedule
generation scheme (SSGS) and the parallel schedule generation scheme (PSGS). These
schemes are described in detail in the following Sections. To overcome their specific lim-
itations, a novel pseudo-serial schedule generation schema was developed in the frame of
this thesis which is presented in Section 4.1. In order to discuss the differences between
the schedules generated by the individual schemes, the different types of schedules are
introduced.
3.4.4.1 Types of schedules
Sprecher et al. (1995) described four different types of schedules, feasible, semi-active,
active, and non-delay schedules2 . In order to illustrate each of the schedule types, the
anchored bored pile wall project example has been modified, in order to clarify the differ-
ences. The resource limit was reduced to three workers, the duration of E12 was reduced
to four time units, and the anchoring activity A2 requires one additional worker. To be
able to understand the differences among the schedule types, two schedule operations
have to be introduced, the local and the global left shift.
A local left shift continuously moves an activity to the left, to an earlier time, without
at any time during the shift creating an unfeasible schedule. In Figure 3.20 shifting activity
E3 from time t2 left towards t1 illustrates such a local left shift. E3 can be placed anywhere
between the two time points without violating any constraints.
A global left shift can pick up any activity and move it to another position that is
earlier in time. At the final position no constraint may be violated, but it is not required
to only pass feasible solutions during the move. In the example moving activity E4 from
time t4 to time t3 illustrates a global left shift. It is feasible to execute the activity at both
points in time, but it cannot be scheduled in between, as in combination with the activity
B2 it would exceed the resource limit.
Figure 3.20: Illustration of global and local left shift operations. Left shifting E3 to t1 is
a local left shift (Labeled l), while shifting E4 to t3 is a global left shift (Labeled g).
2
This Section is based on the description by Sprecher et al. (1995)
Figure 3.21 shows a feasible schedule for the problem. A feasible schedule is any
schedule satisfying all precedence and resource constraints. As is shown in Figure 3.21
the schedule may contain gaps and it may therefore be possible to locally or globally left
shift activities.
Figure 3.21: Feasible schedule for anchored bored pile wall project example.
A semi-active schedule is a schedule were none of the activities can be locally left
shifted. As is illustrated in Figure 3.22 such a schedule may not contain any gaps, but it
may be possible to globally left shift activities, such as activity E4 can be moved to start
right after activity E12. If no activity can be locally or globally left shifted , the schedule
is called an active schedule. An active schedule is not necessarily an optimal schedule.
Figure 3.22: Semi-active schedule for anchored bored pile wall project example.
A non-delay schedule is a schedule where each activity is started at the earliest point
where it becomes resource and precedence feasible. Such a schedule is shown in Figure
3.23. Compared to the schedule in Figure 3.22 the activity E4 becomes precedence and
resource feasible right after the activity E12 finished and therefore has to be scheduled at
that time. A way to verify whether a schedule is a non-delay schedule is to convert the
schedule to a schedule for the corresponding unit-time-duration RCPSP (UTDRCPSP),
where each activity has the exact same duration. For the schedule in Figure 3.23 this is
shown in Figure 3.24. If the original schedule is a non-delay schedule, then the unit-time
transformed schedule is an active schedule (Sprecher et al., 1995).
To show the shortcomings of non-delay schedules, Figure 3.25 shows an active sched-
ule that is much shorter than the non-delay schedule in Figure 3.24. It is visible that there
is a short gap right after activity E12 is finished. At this point activities E3 and E4 are
both precedence and resource feasible, and in a non-delay schedule have to be executed at
that point, creating a suboptimal schedule. It is therefore important to not only consider
non-delay schedules when searching for optimal schedules, as delaying an activity can
Figure 3.23: Non-delay schedule for anchored bored pile wall project example.
Figure 3.24: Unit time duration transformation of the schedule in Figure 3.23
yield an overall shorter schedule. Lemma 4 describes how the different schedule types
are related in terms of set theory.
Figure 3.25: Active schedule for anchored bored pile wall project example.
Lemma 4. Let S denote the set of all schedules, SF the set of feasible schedules, SSA the
set of semi-active schedules, SS the set of active schedules, and SN D the set of non-delay
schedules. Then following holds:
SN D ⊆ SA ⊆ SSA ⊆ SF ⊆ S (Sprecher et al., 1995).
3.4.4.2 Serial Schedule Generation Schemes (SSGS)
For the purpose of generating schedules, different schedule generation schemes exist.
To be able to explain these different scheduling schemes, priority values were assigned
to each activity in the anchored bored pile wall project example as shown in Figure 3.26.
The numbers on the right, inside each of the activities’ box indicates the respective priority
value. The number on the upper right of each box indicate the number of workers required
for the activity. There is a total of three workers available. The priorities were assigned in
an arbitrary way and form the basis for optimization procedures. Logically it makes sense
to have the priorities respect the precedence relationships, because an activity can never
be executed prior to its predecessors. The scheduling schemes select the lowest priority
values first. This means that a lower priority value implies a higher priority. This choice
will be motivated in Section 4.1.
Figure 3.26: Priorities assigned to each activity of the anchored bored pile wall project
example.
Most commonly the serial schedule generation scheme is used, because it can be used
to represent the entire space of active schedules using priority rules (Kolisch, 1996b).
A priority rule is based on a priority value assigned to each activity. Depending on the
priority rule an activity is selected to be scheduled next. For simplicity all priority rules in
this paper select the lowest priority value first. An overview of priority rules is presented
in Section 3.4.8.1.
In the serial schedule generation scheme the priorities define an absolute order in
which the activities are dealt with. The order of the priority values must not violate any
precedence constraints. This means that no activity preceding another activity can have a
higher priority value. Hence the preceding activity is always selected first. The schedule
generation scheme then schedules the activities in the exact order of the priorities, re-
specting resource constraints. If resources are unavailable at the end time of the previous
activity, it is moved forward until the activity blocking the resources has finished. The
procedure is listed in Algorithm 1. The scheme is named serial, as only one activity is
considered for scheduling at a time.
Lemma 5. All active schedules can be generated by the serial schedule generation scheme
(Kolisch, 1996b).
Figure 3.27 shows a schedule generated for the problem with the priorities shown
in Figure 3.26. The bottom part of the figure shows the corresponding resource profile.
Algorithm 1 Serial Schedule Generation Scheme

1: procedure SSGS
E(g): all activities which can be started precedence-feasible in stage g
F(j): maximum finish time of all predecessors of activity j
Li : latest start time of activity i
2: for g = 0 to n + 1 do
3: Calculate the eligible set E(g)
4: Select i ∈ E(g) with lowest priority value
5: Schedule i at the earliest precedence- and resource-feasible start time t ∈
[F(i), Li ]
(Pseudo code from Merkle et al. (2002))
Figure 3.27: Schedule and resource profile created by the serial schedule generation
scheme based on the priorities assigned in Figure 3.26. The critical path is highlighted in
blue.
.
The schedule is generated by treating the activities in the exact order of the priorities as
follows:
1. The activity B1 has the lowest priority value and no predecessors. All three workers
are available, therefore the activity is scheduled at time t0 .
2. E11 has the lowest priority. it is only precedence feasible after B1 has finished. It
is scheduled right after B1 at t1 .
3. A1 requires E11 to be finished. It is scheduled right after E11 at t3 .
4. E21 succeeds A1 and is scheduled right after at t5 .
5. B2 has lowest priority value of the remaining activities. Neither with B1, nor with
E11 does it exceed the resource limit of three workers. Therefore it can be sched-
uled at time t0 .
6. E12 requires B2 to be finished and exceeds the resource limit in combination with
E11. It does however not exceed the limit togethet with A1. The earliest feasible
start time is therefore directly after E11 has finished at t3 .
7. A2 required E12 and is resource feasible in parallel with A1 and E21. It can there-
fore be scheduled directly after E12 at t4 .
8. E22 succeeds A2 and does not exceed the resource limit when combined with E21.
It is scheduled directly after A2 at t6 .
9. E3 does not have predecessors. As it has a longer duration than B1 it can not be
scheduled at time 0. Also it is longer than E12 and therefore does not fit parallel to
that activity either. The earliest slot with one free worker is in parallel with A2 at
t4 .
10. E4 can only be scheduled in parallel to E22 at t6 .
11. As F required E4 to be scheduled, the only possible time is in the end at t8 .
12. No more activities are left to be scheduled.
3.4.4.3 Parallel Schedule Generation Scheme (PSGS)
In contrary to the serial scheme, the parallel schedule generation scheme, also known
as the Brooks Algorithm (BAG), which was first defined by Brooks & White (1965) and
extended by Whitehouse (1979), generates schedules by time increments. At each discrete
time instance, where a previously scheduled activity ends, a list of unscheduled activities
with satisfied resource and precedence constraints is generated, similar to the constraint-
based simulation approach by König & Beißert (2009), as described in Section 2.6.4. The
next activity to be scheduled at this time is selected from this list by a priority rule. The
scheduling procedure is listed in Algorithm 2.
Lemma 6. All schedules generated by the parallel schedule generation scheme are non-
delay schedules (Kolisch, 1996b).
Figure 3.28 shows a schedule generated for the problem with the priorities shown
in Figure 3.26. The bottom part of the figure shows the corresponding resource profile.
The schedule is generated by discrete time increments and selecting currently feasible
activities by choosing the activity with the lowest priority value:
Algorithm 2 Parallel Schedule Generation Scheme

1: procedure PSGS
J : set of all activities
Er (g): all activities which can be started resource- and precedence-feasible in stage g
C: set of already scheduled activities
Ag : set of active activities
2: g := 0, tg := 0, C := ∅
3: while |J − C| > 0 do
4: Calculate the eligible set E(tg )
5: while E(tg ) 6= ∅ do
6: Select i ∈ E(tg ) with lowest priority value
7: Schedule i at tg
8: C := C ∪ {i}
9: recalculate E(tg )
10: Calculate Ag
11: g := g + 1
12: Calculate the minimal finish time tg of all activities in Ag−1
(Pseudo code from Merkle et al. (2002))
1. The procedure starts at time t0 .

At this point activities B1, B2, E3, and E4 are precedence and resource feasible. B1
has the lowest priority value and is therefore scheduled first.
Now B2, E3, and E4 are still feasible. Therefore B2, as it has lowest priority, is
scheduled. Since E3 and E4 are still feasible, E3 is scheduled next. Now no more
activities are feasible as all resources are in use.
2. The next activity that finishes is B1. Therefore time is incremented to t1 .

The only feasible activity at this time is E4 which is scheduled next.
3. The next finish time is t2 .

Now E11 and E12 are feasible. As E11 has a lower priority value, it is scheduled
next. No more activities are feasible at this time.
4. Both E4 and E11 end at time t3 .

Activities A1 and E12 are feasible now. As A1 has a lower priority value, it is
scheduled next. Since E12 is still feasible, it is also scheduled.
5. Time is incremented to the end time of E12, t4 .

The only feasible activity is A2 which is scheduled.
6. A1 ends at time t5 .
E21 is the only feasible activity and is therefore scheduled.
7. A2 ends at time t6 .
E22 is now feasible and scheduled.
Figure 3.28: Schedule and resource profile created by the parallel schedule generation
scheme based on the priorities assigned in Figure 3.26. The critical path is highlighted in
blue.
.
8. E21 finished at time t7 .

The only remaining activity F is not yet precedence feasible.
9. E22 finishes at time t8 .

The last activity F is now feasible and scheduled.
10. F finishes at time tp .

No more activities are left.
3.4.4.4 Differences between the Schedule Generation Schemes
When comparing the results of both schedule generation schemes in Figures 3.27 and
3.28, it can be observed that the main difference lies in the gaps within the resource
profile of the serial schedule between times t0 and t1 , and t2 and t3 . Those gaps cannot be
created in the parallel scheme as activities E3 and E4 are used to fill the gaps. This also
gives rise to the naming of the parallel scheme, as it maximizes parallel execution in a
greedy way. Therefore the parallel schedule generation scheme can only create non-delay
schedules, as is stated in Lemma 6. It can also be observed that both schedules form a
different critical path, which is highlighted in blue in both schedules, and therefore have
different makespans.
Generally the serial schedule generation scheme has higher computational demands
because it needs to search for the earliest precedence- and resource-feasible start time for
every activity, while the parallel scheme performs time increments. However, the run time
complexity for both schemes is O(N 2 · |R|) (Kolisch & Hartmann, 1999; Pinson et al.,
1994), since in the worst case each activity has to be started at an individual time instance.
In practical cases, the runtime of the serial scheme is much higher, as is illustrated in Fig-
ure 3.29. It shows the simulation run time of both scheduling schemes for 1000 randomly
generated problems consisting of up to 1000 activities.
400
Parallel
Serial
300
Simulation run time in ms
200
100
0
0 200 400 600 800 1000
Number of activities
Figure 3.29: Comparison of performance of the serial and parallel schedule generation
schemes with random problems with up to 1000 activities.
.
The author of this thesis has developed the new pseudo-serial schedule generation
scheme, which is presented in Section 4.1. It combines the advantages of both the serial
and the parallel schedule generation schemes.
3.4.5 Scheduling using Simulation
Simulation based scheduling techniques were initially applied to scheduling problems

in the manufacturing industry (Jackson, 1957). The related problem is called the job
shop scheduling problem, and it consists of a set of jobs and a set of machines which
perform operations on jobs. Each job must go through a specific order of the machines
in order to be completed. There are no precedence constraints among the operations
of different jobs, operations cannot be interrupted, and each machine can only operate
on one job at a time, as well as one job can only be operated on by one machine at a
time. The objective is to find job sequences that minimize the makespan of all operations
(Błażewicz et al., 1996). Same as the resource constrained project scheduling problem,
the job shop scheduling problem is NP-hard (Lenstra & Rinnooy Kan, 1979). The main
difference between the problems is, that the job shop scheduling problem involves a set of
stationary machines where jobs pass through, whereas the RCPSP involves one large job
consisting of many activities (operations) that need to be performed involving resources
(machines). Furthermore, in the RCPSP, multiple resources can be required for a single
activity. Kiran & Smith (1984) presented an overview of simulation based methods for job
shop scheduling. For RCPSPs the simulation-based approaches merely use the simulation
model to generate schedules based on predetermined parameters, such as priority values,
that are set by a heuristic algorithm, such as those described in Section 3.4.8.
The first simulation model that was widely applied to construction operations was
CYCLONE, which is based on the modeling approach presented in Section 2.5.4 (Halpin,
1977). Later STROBOSCOPE was developed, which is based on a similar modeling
approach, which is presented in Section 2.5.5 (Martinez & Ioannou, 1994). The ABC
approach of Shi (1999) was presented in Section 2.5.6. All those approaches focus on
cyclic operations, such as earthworks, or the bored drill pine example. Using them as a
model for an entire construction project causes the model to become highly complex.
Simulation based methods for RCPSPs only got into focus of research as sufficient
computational power for simulating large construction projects became available. König
et al. (2007) introduced the constraint-based simulation approach in order to create sched-
ules for civil engineering projects and ship construction. The constraint-based simulation
approach has strong similarity to the parallel schedule generation scheme presented in
Section 3.4.4.3, but features some key differences. The main difference is that the next
activity to be scheduled is selected randomly. Through multiple simulation runs in a
Monte Carlo way (Section 2.2.6), well-performing schedules are generated. Additionally,
incorporating soft-constraints allows to include experiences from similar projects, human
strain factors, and preferences of the planners (Beißert et al., 2010). As these constraints
do not strictly need to be satisfied, they are modeled as soft-constraints. The simulation
concept is illustrated in Figure 3.30.
Additionally König’s simulation concept involves productivity of workers and ma-
chines under different circumstances, that may influence the durations of activities, as for
instance, having to perform an activity in a confined space may reduce efficiency. Further-
more the concept was extended by incorporating uncertainty in order to create schedules
which maximize robustness (König, 2011a,b). Moreover the feedback of actual progress
data from construction sites was enabled by Szczesny & König (2015). It involves updat-
ing a given schedule based on real-time data, and if considered necessary, create a new
schedule using evolutionary algorithms in combination with König’s simulation approach.
Updating may be required due to extensive delays of the involved activities, which is dis-
cussed in more detail in Section 4.6.
Figure 3.30: Flowchart of simulation concept by Beißert et al. (2010).
3.4.6 Simulation-based Critical Path Analysis using the Dori-Algorithm
In order to know when a delay is causing a schedule update to be necessary it is useful

to know both the critical path of the schedule and the float time of each activity as it is
described in Section 3.3.3 about the critical path method (CPM).
The CPM however does not take resource limits into account. This limitation was ad-
dressed by Dori et al. (2012) and Dori (2016) who published a simulation based algorithm
to determine both float times and the critical path of a given RCPSP schedule.
Similar to the critical path method, the Dori-Algorithm (DA) is a two pass procedure
based on a forward and a backward simulation run. However, due to resource limitations
a direct backward simulation run can yield a different order of activities using the same
resources, and therefore result in incorrect float times. In order to fix this, additional
constraints need to be introduced for the backward simulation run. These constraints are
required for every set of activities that are not linked by precedence constraints, but in
combination exceed the number of available resources. This is referred to as a Dori-Set
(DS). A Dori-Set is a special case of a forbidden set, as defined by Schäffter (1997). It is
to be ensured, that the backward simulation run does not create switches in the order of
activities that cannot be scheduled simultaneously due to resource constraints.
As an example, the schedule for the anchored bored drill pine project example, gen-
erated by the serial schedule generation scheme, which is shown in Figure 3.27 is used.
It is assumed that this is the result of the forward simulation run. The example involves
only one type of resource and therefore the Dori-Sets can be determined easily. B1 and B2
are not precedence linked, but together use a total number of two resource instances. The
next finishing independent activity is E12 and the combination of those three activities
exceed the number of available resources. The Dori-Algorithm now dictates an artificial
sequence enforcement constraint to be added between B1 and all direct successors of E12,
which is only the activity A2. Then E11 and E12 form a Dori-Set and therefore a sequence
enforcement constraint between E11 and A2 is required. Another Dori-Set is formed by

E12, A1, and E21, which adds a sequence enforcement constraint between E12 and F. All
other Dori-Sets involve F which does not have successors. Therefore no other constraints
need to be added. Figure 3.31 shows the schedule with the added constraints as thick
red arrows. The result of the backward simulation run is depicted in Figure 3.32, and
the resulting critical path and float times are shown in Figure 3.33. For a more in-depth
description of the Dori-Algorithm the reader is referred to Dori (2016, pp 103–116).
Figure 3.31: Schedule from Figure 3.27 with sequence enforcement constraints added by
the Dori-Algorithm depicted as thick red arrows.
.
Figure 3.32: Schedule resulting from backward simulation run according to the Dori-
Algorithm based on the sequence enforcement constraints in Figure 3.31.
.
Figure 3.33: Float times and critical path resulting from Dori-Algorithm for the schedule
in Figure 3.31.
.
3.4.7 Deterministic Approaches to the Resource Constrained Project Scheduling

Problem
Since the resource constrained project scheduling problem is NP-Hard in the strong sense
(Blazewicz et al., 1983), deterministic approaches are hardly feasible for large real-life
problems. However, for completeness, the theoretical approaches for deterministic RCPSP
solutions are described within this Section3 . Talbot & Patterson (1978) defined a con-
ceptual linear program in order to solve the project scheduling problem without taking
resource constraints into account. This linear program is shown in Equation 3.12. This
definition is forms a basis for the definitions of the RCPSP and is therefore described at
this point. It introduces two artificial activities called dummy activities 0 and J + 1. Ac-
tivity 0 is the start activity and it precedes all other activities while activity J + 1 is the
final activity which succeeds all other activities. Both dummy activities have a duration of
0. The objective of the linear program is to minimize the starting time FJ+1 of the finish
activity as is shown in Equation 3.12a. Equation 3.12b defines all precedence constraints
Vij . It defines that the finish time of activity i, Fi must be smaller or equal to the finish
time Fj of activity j, minus the duration dj of activity j, given that there is a precedence
constraint that defines activity j to be a successor of activity i. Equation 3.12c defines the
start activity to finish at time 0, while Equation 3.12d defines the finish times to be larger
or equal than 0 for the set of all activities J . Equation 3.12e defines that all finish times
must be smaller or equal to some predetermined project horizon, as for instance the sum
of all activities’ durations.
minimize FJ+1 (3.12a)

subject to Fi ≤ Fj − dj , ∀hi, ji ∈ Vij (3.12b)
F0 = 0 (3.12c)
Fi ≥ 0, ∀i ∈ J (3.12d)
FJ+1 ≤ d¯ (3.12e)
Pritsker et al. (1969) however defined a different approach to solve PSPs using linear
programs involving time indexing. They therefore introduced a binary variable xit as
defined in Equation 3.13 that is 1 if and only if activity i starts at time t. Ei is the set
of possible start times of activity i. This set can be determined for each activity using
the critical path method as described in Section 3.3.3. The sum defined in Equation 3.14
therefore results in the time Fi , where activity i finishes, as only one xit can be 1 for
each activity i, as is defined in the constraint defined in Equation 3.15b. The objective
function is equal to the finishing time of the final dummy activity as defined in Equation
3.15a. Equation 3.15b defines the precedence constraints equally to Equation 3.12a after
substitution according to Equation 3.14. Equation 3.15d defines that the initial dummy
activity start at time 0 and Equation 3.15e defines that all xit are binary variables.
3
This Section was inspired by the description by Bauer (2009)
(
1 if t = Fi
xit = (3.13)
0 otherwise
X
Fi = t · xit (3.14)
t∈Ei
X
minimize t · xJ+1,t (3.15a)
t∈EJ+1
X
subject to xit = 1 ∀i ∈ J (3.15b)
t∈Ei
X X
t · xjt − t · xit ≥ dj ∀hi, ji ∈ Vij (3.15c)
t∈Ej t∈Ei
x00 = 1 (3.15d)
xit ∈ {0, 1} ∀i ∈ J ; ∀t ∈ Ei (3.15e)
This time indexed notation allows the definition of an additional constraint modeling
resource limitations as defined in Equation 3.17d. R is the set of all resources and T
is the set of all time periods. kr is the limit of resource r and kir is the consumption of
resource r for activity i per time period. Equation 3.16 defines kir as the overall demand
uir of resource r by activity i divided by the duration di of activity i. Qit is the set of time
periods where activity i is in progress if it was started at time t. Therefore the constraint
in Equation 3.17d prevents any resource limit to be exceeded at any time.
uir
kir = (3.16)
di
X
minimize t · xJ+1,t (3.17a)
t∈EJ+1
X
subject to xit = 1 ∀i ∈ J (3.17b)
t∈Ei
X X
t · xjt − t · xit ≥ dj ∀hi, ji ∈ Vij (3.17c)
t∈Ej t∈Ei
X X
kir xiτ ≤ kr ∀r ∈ R; ∀t ∈ T (3.17d)
i∈J τ ∈Qit
x00 = 1 (3.17e)
xit ∈ {0, 1} ∀i ∈ J ; ∀t ∈ Ei (3.17f)
When examining the linear program defined by Equation 3.17 several downsides be-
come clear. For activities with very long durations the set T can become very large. This
can be counteracted by setting the length of a time period to the greatest common divi-
sor of all activities’ durations. However, for instance two activities with durations 1 and
10000 would still require 10000 time periods to be defined. Each time period in T cre-
ates one constraint for each resource in R and each activity in J creates one variable xit
for each possible start time in Ei . This causes the linear program for larger problems to
become extremely large. This illustrates why heuristic procedures are of high interest for
research.
3.4.8 Heuristic Approaches to the Resource Constrained Project Scheduling Prob-

lem
As shown in the previous Section, deterministic solutions for large scale problems are
hardly feasible, and therefore heuristic procedures are highly interesting for practical
problems. This Section provides an overview of the state-of-the-art of heuristics for
resource-constrained project scheduling problems4 .
3.4.8.1 Priority Rules
A priority rule assigns a value v(i) to each activity i and states whether the minimum
or maximum value is to be selected first (Kolisch & Hartmann, 1999). Additionally a
way to handle ties needs to be defined. The priority rules can be used by either of the
schedule generation schemes defined in Section 3.4.4 but could also be incorporated into
the constraint-based discrete-event simulation approach by Beißert et al. (2007), which
is described in Section 2.6.4. Özdamar & Ulusoy (1995) defines the difference between
the scheduling schemes as the employment of either fixed a-priori priorities (serial) or
dynamically updating priorities (parallel). In this thesis priorities are always assumed to
be set a-priori, but for the serial schedule generation scheme need to be in precedence
feasible order.
The priorities can either be assigned randomly and used as a basis for optimization
procedures, or be calculated based on different criteria. The individual rules can be clas-
sified based on the information used to calculate the priority values v(i), which can be
network, time, and resources, as well as lower and upper bounds (Kolisch & Hartmann,
1999). For an overview of many different priority rules the reader is referred to the papers
by Davis & Patterson (1975), Kolisch (1996a), and Hartmann & Kolisch (2000).
3.4.8.2 X-Pass Methods
X-Pass methods deploy different priority rules in combination with different schedule
generation schemes on the problem. The simplest way are single pass methods which
employ exactly one priority rule in combination with one schedule generation scheme.
Multi-pass methods can be created by different combinations of schedule generation
4
This Section was in part inspired by the review papers by Hartmann & Kolisch (2000); Kolisch &
Hartmann (1999, 2006).
schemes and priority rules, or by forward-backward scheduling methods. Another cat-

egory of x-pass methods are sampling methods which make use of randomly selected
priority rules, randomly generated priority values, or priority rules based on selection
probabilities. Among the most advanced versions is adaptive regret based biased random
sampling (RBRS) as presented by Kolisch & Drexl (1996) and Schirmer (2000). The idea
is to select the priority rule, schedule generation scheme and the sampling method based
on problem characteristics, as described in Section 3.4.2, such as number of activities or
resource constrainedness.
3.4.8.3 Tabu Search
Glover (1989, 1990a,b) introduced the method of tabu search which can be applied to
resource constrained project scheduling methods. It is basically a steepest decent neigh-
borhood search. Starting from an initial solution, neighboring solutions are analyzed for
the best solutions. As it does not strictly require improving solutions, it is possible that
the algorithm starts to cycle. To prevent this, Glover introduced a kind of memory, re-
membering previously encountered solutions to prevent going back to a recently visited
solution. This memory can be depleting, such that only a certain number of previously
visited solutions are stored in order to save memory and computational effort. An appli-
cation of tabu search for resource constrained project scheduling was published by Baar
et al. (1999).
3.4.8.4 Simulated Annealing
Kirkpatrick et al. (1983) introduced the algorithm called simulated annealing, combining
combinatorial optimization and statistical mechanics into a new metaheuristic optimiza-
tion algorithm. It is based on the process of the annealing of metals where free floating
particles cool down and lock into a state of low energy. After an initial solution is gen-
erated, the neighborhood is analyzed for other solutions. If a better solution is found it is
always accepted as the new current optimum. If no better solution is found, but a worse
one, it is only accepted with a probability depending on the quality of the solution and the
current temperature of the simulated cooling matter. The temperature Tt is lowered during
the execution of the algorithm until it is cooled down, according to a cooling scheme. The
probability P (y|x) of accepting a new solution y, given the current solution x is defined
in Equation 3.18, where f (x) is the fitness of solution x.
(
1 if f (y) ≤ f (x)
P (y|x) = (3.18)
exp(− f (y)−f
Tt
(x)
) if f (y) > f (x)
Simulated annealing was applied to resource constrained project scheduling problems

by Józefowska et al. (2001), Bouleimen & Lecocq (2003), Valls et al. (2005), and König
& Beißert (2009).
3.4.8.5 Genetic Algorithms
Holland (1975) introduced a new nature inspired metaheuristic algorithm which was in-
spired by adaption and evolution in biology. While the simulated annealing algorithm
works on improving a single solution, a genetic algorithm works on a population of solu-
tions which are improved using a set of operators which either modify (mutate) a member
of the population or mate (crossover) two members of the population. The population is
usually kept at a certain number of members. When through modification that number
is exceeded, the weakest members of the population die out. Weakness in this context
is commonly determined by a fitness function. In the scheduling domain, the weakest
solutions would be those having the longest makespan. Genetic algorithms were applied
to resource constrained project scheduling problems by Hartmann (1998), Toklu (2002),
Valls & Ballestín (2002), Senouci & Eldin (2004), Debels & Vanhoucke (2005), Debels
et al. (2006), Valls et al. (2008), Long & Ohsato (2009), König (2011a), and Ghoddousi
et al. (2013).
3.4.8.6 Ant Colony Optimization
Dorigo et al. (1996) introduced the ant colony optimization algorithm as a new meta-
heuristic. The idea of the algorithm is to simulate a population of agents driven by a
greedy force. Through the interaction of the agents by so-called pheromones, the per-
formance of the algorithm was improved significantly over the performance of a single
agent. The algorithm was adapted to resource-constrained project scheduling problems
by Merkle et al. (2002). They used the serial schedule generation scheme and pheromones
reflect the expected efficiency of putting certain activities in specific places using the lat-
est start time priority rule, where activities are ordered by their latest start time according
to the critical path method described in Section 3.3.3.
3.4.8.7 Local Search Procedures
Local search is a procedure that starts from random points in the search space and tries to
find improved solutions by local modifications of the solutions. In the case of scheduling
this is usually done by modifying the priority values. These methods include procedures
such as the previously introduced simulated annealing and tabu search methods. Simu-
lated annealing is a stochastic local search procedure as defined by Hoos & Stützle (2004).
Fleszar & Hindi (2004) use a variable version of neighborhood search applied to RCPSPs
based on the serial schedule generation scheme. Valls et al. (2003) uses a topological or-
der representation of the schedules encountered, combining local search with tabu search,
while also using the serial schedule generation scheme.
3.4.8.8 Swap-based Local Search
Another possibility for local search algorithms is the use of swaps between two activi-
ties, that are precedence feasible, modifying an existing schedule (Bügler & Borrmann,
2014; Bügler et al., 2013; Dori, 2016). A swap is commonly performed by interchanging
the priorities of two activities. A special kind of swap, a cyclical shift, was used in the
simulated annealing algorithm by Bouleimen & Lecocq (2003). The effect of a normal
swap differs depending on the schedule generation scheme that is used. The schedule
generation schemes were introduced in Section 3.4.4. In order to illustrate the effects of
swapping activities a simplified version of the bored pile wall project example is intro-
duced in Figure 3.34. It features two bored pile walls B1 and B2, both requiring a drilling
rig, which is colored in yellow. Furthermore two excavation activities E1 and E2 need to
be scheduled, requiring an excavator, which is colored in green. Finally a foundation can
be build, which is not assigned any resources in this example. There is only one drilling
rig and one excavator available, hence neither B1 and B2, nor E1 and E2 can be executed
in parallel.
Figure 3.34: Simplified bored pile wall project example with priorities on the right inside
of each activity’s box.
Figure 3.35 shows an example schedule for the simplified bored pile wall project
example in Figure 3.34. The schedule could have been create by either of the two schedule
generation schemes based on the priorities defined a-priori.
Figure 3.35: Example schedule for simplified bored pile wall project example in Figure
3.34.
Figure 3.36 shows the schedule after E1 and E2 have been swapped. This swap how-
ever only has an effect using the serial schedule generation scheme. When activity B1
ends, E1 immediately becomes resource and precedence feasible and therefore is sched-
uled at this very time instance. In the serial scheme however, E2 is scheduled at the
earliest resource and precedence feasible time, right after B2 finished, leaving insufficient
room for E1 and therefore pushing it to start after E2. This creates a longer schedule as
the initial one.
Figure 3.36: Example schedule from Figure 3.35 after swapping E1 and E2 using the
serial schedule generation scheme.
When an additional swap is performed, switching the priorities of B1 and B2, this
results in the schedule shown in Figure 3.37. This schedule would, in this priority config-
uration be equal for both scheduling schemes. The resulting schedule is shorter than the
initial one and is actually an optimal solution for the problem.
Figure 3.37: Schedule from Figure 3.35 after additionally swapping B1 and B2.
When swapping B1 and B2 first, however, using the serial scheduling scheme, a longer
schedule is created, similar to the one in Figure 3.36. This is shown in Figure 3.38.
When on the other hand the parallel scheme is used for the exact same swap, the
schedule collapses into the optimal solution right away, as is shown in Figure 3.39. This
is due to the property of the parallel schedule generation scheme stating that it only creates
Figure 3.38: Example schedule from Figure 3.34 after swapping B1 and B2 using the
serial schedule generation scheme.
non-delay schedules and therefore minimizes idle times in a greedy manner. This however
is not always optimal as is shown in the following example.
Figure 3.39: Example schedule from Figure 3.34 after swapping B1 and B2 using the
parallel schedule generation scheme.
When viewing another modification of the example, where instead of the second bored
pile wall, a sheet pile wall is created, and therefore a piledriver is required for the activity
P1. The modified problem, with activity priorities, is shown in Figure 3.40.
A schedule, resulting from the priorities in Figure 3.40, is shown in Figure 3.41. This
schedule would result from both scheduling schemes, given the assigned priorities.
However, wehen E1 and E2 are swapped, this does not have any effect on the result
of the parallel schedule generation scheme, as E2 is the only precedence and resource
feasible activity at t1 , when P1 is finished. In the serial schedule generation scheme on
the other hand, E1 will be scheduled at time t2 , right after B1 has finished, and in the
result create a much shorter schedule, as shown in Figure 3.42. Due to the unnecessary
idle time of the excavator between times t1 and t2 , this is not a non-delay schedule and
therefore can in no way be generated by the parallel schedule generation scheme.
Figure 3.40: Simplified bored pile wall and sheet pile wall project example with priorities
on the right inside of each activity’s box.
Figure 3.41: Schedule for the example in Figure 3.40.
Figure 3.42: Schedule for the example in Figure 3.40 after swapping E1 and E2, when
using the serial schedule generation scheme.
Lemma 7. Every active schedule can be reached from any other schedule by applying a
series of swaps using the serial schedule generation scheme.
Lemma 8. Every non-delay schedule can be reached from any other schedule by applying
a series of swaps using the parallel schedule generation scheme.
Dori (2016) defined the notion of reasonable swaps, as swaps between activities which
share the usage of a common resource, and cause an actual change in the resulting sched-
ule. Whether a swap results in a different schedule depends on the scheduling scheme
that is used. As the work of Dori (2016) builds up on the constraint-based simulation by
König et al. (2007), the definition of reasonable swaps is based on the parallel schedule
generation scheme as defined in Section 3.4.4.3.
The parallel schedule generation scheme allows the collection of such swaps during
schedule generation with minimal effort. Every time during the schedule generation pro-
cess, the eligible set E(g) is updated, after an activity was scheduled. When due to the
scheduled activity, activities previously in the eligible set, were removed, this must be
caused by resources shared by the executed activity and the removed one, as the prece-
dence feasibility cannot have changed at this point. Therefore any swap between the
scheduled activity and any subsequently removed activity is reasonable according to the
definition by Dori (2016).
The notion of reasonable swaps can be used as a basis for several metaheuristic algo-
rithms. Dori (2016) calculated the total possible number of swaps Ψ as defined in Equa-
tion 3.19 based on the number of activities N . This results in a total possible number of
resulting schedules |S| as defined in Equation 3.20.
N2 − N
Ψ= (3.19)
2
Ψ
X 1
|S| = Ψ! · (3.20)
i=0
i!
It is obvious that the worst-case number of possible schedules is enormous. For a

problem with N = 10 activities, a maximum of Ψ = 45 swaps is possible. This results
in a maximum number of schedules |S| = 3.2517 · 1056 . In order to not have to ex-
plore all those possible schedules, Dori (2016) introduces two techniques for intelligently
exploring only parts of the search space by utilizing reasonable swaps. First a simu-
lated annealing algorithm, similar to the one proposed by Bouleimen & Lecocq (2003),
which is described in Section 3.4.8.4, but instead of cyclical shifts, it is using reason-
able swaps. Another approach is the creation of search trees by swaps as is illustrated in
Figure 3.43. Each node in this tree represents one schedule by a vector of priorities. A
swap is performed by interchanging priorities within this vector. As the set of reasonable
swaps changes for each schedule, it has to be recalculated for each node. Dori (2016) pro-
posed a greedy-like heuristic search on such a tree, by exploring the search tree by only
investigating nodes improving the current solution, but having a certain fixed tolerance
towards worse solutions in order to overcome local optima. This approach is similar to
the simulated annealing approach, but it is accepting solutions in a non-stochastic manner,
by a fixed threshold. Furthermore it searches along multiple branches in a breadth-first

manner, which additionally facilitates parallel execution of the algorithm.
Figure 3.43: Tree created from swaps for a problem with 5 activities. Each node contains
a vector of priorities. One pair of priorities is swapped along each arc.
In summary Dori (2016) has shown that swap-based local search algorithms are an
effective heuristic search procedure for RCPSPs. A downside is the humongous size of
the resulting search trees, making the search within the tree very tedious. This issue was
addressed by the author of this thesis when developing the iterative swap-based limited
depth search algorithm, that is presented in Section 4.3.
3.4.8.9 Overview of Heuristic Procedures for RCPSPs
This Section provides a tabular overview of the different published heuristic algorithms.
The algorithms are classified into the previously described categories and sorted by publi-
cation date. Additionally the table includes many review papers, that provide comparisons
between the different published algorithms.
Algorithm class References

Genetic Algorithms Chan et al. (1996), Hartmann (1998), Alcaraz & Maroto
(2001), Toklu (2002), Valls & Ballestín (2002), Kochetov &
Stolyar (2003), Senouci & Eldin (2004), Valls et al. (2004),
Kim et al. (2005),Debels & Vanhoucke (2005), Debels
et al. (2006), Jaśkowski & Sobotka (2006), Gonçalves et al.
(2008), Valls et al. (2008), Alcaraz & Maroto (2009), Long
& Ohsato (2009), Agarwal et al. (2011), Ballestín et al.
(2011), König (2011a), Ghoddousi et al. (2013)
Hybrid Algorithms Drexl (1991), Demeulemeester & Herroelen (1997), Mika
et al. (2005), Rivera (2013)
Particle Swarm Opti- Zhang et al. (2005a), Zhang et al. (2006), Lu et al. (2008),
mization Chen et al. (2010), Koulinas et al. (2014)
Priority Rules Thesen (1976), Kurtulus & Davis (1982), Kolisch (1996a),
Klein (2000), Browning & Yassine (2010)
Simulated Annealing Boctor (1996), Józefowska et al. (2001), Bouleimen &
Lecocq (2003), König & Beißert (2009), Bettemir & Son-
mez (2012)
Tabu Search Thomas & Salhi (1998), Baar et al. (1999), Waligóra (2009)
Tree Search Simpson & Patterson (1996), Bügler et al. (2013), Bügler &
Borrmann (2014), Dori (2016)
X-Pass Methods Bell & Han (1991), Kolisch & Drexl (1996), Schirmer
(2000), Chtourou & Haouari (2008)
Others Bell & Park (1990), Li & Willis (1992), Leon & Balakr-
ishnan (1995), Salewski et al. (1997), Schirmer (2001),
Tormos & Lova (2001), Calhoun et al. (2002), Merkle
et al. (2002), Valls et al. (2003), Fleszar & Hindi (2004),
Debels et al. (2006), Alvarez-Valdes et al. (2008), Gon-
charov (2011), Horenburg et al. (2012), Aziz et al. (2014),
Jedrzejowicz & Ratajczak-Ropel (2014), Mejía et al. (2016)
Reviews Wiest (1967), Davis & Patterson (1975), Cooper (1976),
Elsayed (1982), Boctor (1990), Boctor (1993), Kolisch
(1996b), Kolisch & Hartmann (1999), Böttcher & Drexl
(1999), Hartmann & Kolisch (2000), Kolisch & Hartmann
(2006), Bauer (2009), Van Peteghem & Vanhoucke (2014)
Table 3.5: List of heuristic optimization procedures for the resource constrained project
scheduling problem.
3.4.9 Benchmarks
In order to evaluate the efficiency of an approach to solve resource-constrained project

scheduling problems, different problem sets have been published. Davis & Patterson
(1975) and Talbot & Patterson (1978) published the first such sets, consisting of 83 and
50 test problems. Later Patterson (1984) created a very popular set consisting of 110
problems of sizes varying between 7 and 50 activities, that was used throughout many
publications. Kolisch & Sprecher (1997) later published the most commonly used prob-
lem sets, the Project Scheduling Problem LIBrary (PSPLIB) (Kolisch & Sprecher, 1996).
It is structured by activity count into four sets with 30, 60, 90 and 120 problems respec-
tively. The 30, 60 and 90 activity sets each contain 480 problem instances and the 120
activity set contains 600 problems. A majority of papers published in the area of RCPSPs
since then used the PSPLIB in order to compare the performance of algorithms. For the
30 activity problems all optimal solutions are known. For all other sets the best heuristic
solutions that have yet been found are stored in the library, as well as lower and upper
bounds for the optimal makespan. For the 60 activity problems, most of the bounds are
closed, such that the optimal makespan is known. The PSPLIB problem sets have been
used in nearly all recent publications on algorithms for the RCPSP, such that it forms a
solid basis for a comparison of new developments with the state-of-the-art. The author
of this thesis has developed an extension of the PSPLIB which is presented in detail in
Section 4.5.
3.5 Summary
Construction schedule optimization forms a vital part of the work presented in this the-
sis because proper reaction to changing circumstances requires optimization procedures
to analyze the space of feasible schedules for good solutions. This Chapter described in
detail, the state-of-the-art in construction schedule optimization. First the possible objec-
tives and schedule metrics were introduced. Additionally the introduction of uncertainty
and robustness into the problems was discussed. As the objectives, including robustness,
are often antithetic or create trade-off problems, the concept of multi-criteria optimization
was introduced. Through the project scheduling problem, which does not take resources
into account, the resource constrained project scheduling problem was introduced. Based
on an illustrative example, the different schedule generation schemes were discussed.
Problem characteristics were compared and simulation-based approaches, such as the
constraint-based discrete-event simulation method and the Dori algorithm were intro-
duced. A detailed overview of heuristic approaches from literature and the introduction
of the commonly used benchmarks were discussed. This Chapter illustrated that there is
a very large basis of methods for solving RCPSPs in literature, but also pointed out some
research voids that are addressed in the following Chapter of this thesis.
97
4 Improved Methods for

Construction Scheduling
"In preparing for battle, I have always found that plans are useless but plan-
ning is indispensable." - Dwight D. Eisenhower
The main contributions presented in this Chapter consist of a new schedule generation
scheme for resource constrained project scheduling problems, the adaption of the water
wave optimization algorithm by Zheng (2015) to this problem, and a swap-based iterative
limited depth search algorithm. The new algorithms offer new tools for planners to gener-
ate near-optimal schedules. Additionally the methods can be used to automate the process
of updating schedules based on newly obtained data in a reactive manner. Furthermore
the author developed a uncertainty extension of the project scheduling problem library
PSPLIB. This extension allows scheduling algorithms to be compared on problems in-
volving uncertain performance factors or activity durations. Such problems are common
when modeling real-life construction sites and uncertainty is often generated as a product
of data acquisition when maintaining a reactive simulation model. This Chapter discusses
each of these contributions in the respective Sections.
4.1 Pseudo-Serial Schedule Generation Scheme

Section 3.4.4 introduced the two most commonly used serial and parallel schedule gen-
eration schemes. Schedule generation schemes form an essential part of many heuristic
algorithms for resource constrained project scheduling problems. While the serial sched-
ule generation scheme can generate all active schedules, as described in Section 3.4.4.1,
it is much slower than the parrallel schedule generation scheme.
As Lemma 4 describes, the set of active schedules always contains an optimal sched-
ule for a given RCPSP. However, in order to define a schedule to be generated by the serial
schedule generation scheme, each activity needs to have a priority value that respects the
precedence relationships, as the activities are processed in the exact order defined by the
priorities. This increases the difficulty of systematically searching the space of possible
solutions, as only a small subset of priority vectors can be used. Zhang et al. (2005a) uses
a method to adjust the vectors to feasible ones prior to schedule generation by sorting the
priority values in order to not violate the precedence constraints according to the work
98 4 Improved Methods for Construction Scheduling
by Chan et al. (1996). This however populates the search space with numerous duplicate
solutions for different input vectors and reduces the continuity of the search space.
The parallel schedule generation scheme does not suffer from this discrepancy, but
does on the other hand not allow the generation of all active schedules, but only non-delay
schedules. Those schedules are formed by time increments, selecting feasible activities
by assigned priority values. Each activity is then started at the earliest possible time.
Because the activities, to be executed next, are selected from only the feasible activities,
as dictated by both the precedence, as well as the resource constraints, the priorities do
not have to reflect the precedence relationships. It has been shown by Kolisch (1996b)
that less than 60% of the problems in the PSPLIB j30 problem set have optimal solutions
that are non-delay schedules, which can be generated by the parallel schedule generation
scheme, as defined in Section 3.4.4.3.
When using schedule generation schemes as a basis for heuristic algorithms, it is de-
sired to have a parameter space that is continuous and contains the optimal solution. This
allows the algorithms to operate by applying arbitrary priority values to activities. Nei-
ther the serial, nor the parallel schedule generation scheme fulfill both criteria. For this
purpose a new schedule generation scheme was developed. The pseudo-serial schedule
generation scheme, yielding a complete and continuous representation of all active sched-
ules.
As the name suggests, the set of non-delay schedules do not include schedules where
an activity is executed later than the earliest point in time where the activity becomes
resource- and precedence-feasible. Hence, no activity can be delayed beyond this point.
The pseudo-serial schedule generation scheme is based on the parallel schedule genera-
tion scheme, extended by the concept of introducing delays. Delays are introduced into
the space of possibilities by first scaling the search space down to a range of priorities in
the interval [0, 1]. This does not have any effect on the resulting schedules as any given set
of priority values can be scaled to fit into that interval without changing the relative order.
Delays are then represented as integer values added to those priorities. When selecting
the activity with the lowest priority value Pj , its fractional part Rj is used instead. An
activity being delayed once, implies that it is ignored when it occurs as the next activity to
be scheduled for the first time. In this representation such an activity would be given an
interval in the range [1, 2]. After it was ignored once, the priority value is reduced to the
range [0, 1] by subtracting 1. When the same activity is selected for scheduling the next
time, it will no longer be ignored.
The entire procedure is summarized in Algorithm 3. Figure 4.1 illustrates the resulting
search space for a problem with two activities. The bottom left rectangle, where both
values V (A) and V (B) are in the [0, 1] interval contains all non-delay schedules that can
be generated by the parallel schedule generation scheme. The areas extending outwards in
the top and left directions contain schedules where at least one activity has been delayed.
The search space is generally bound to positive values.
Figure 4.2 shows an example problem that was introduced in Section 3.4.8.8 involving
five activities and three resources. As discussed in that Section, the only schedule that
can be generated by the parallel schedule generation scheme for this problem is the one
illustrated in Figure 4.3. Activity E2 can executed as soon as activity P1 has finished. As
4.1 Pseudo-Serial Schedule Generation Scheme 99
Algorithm 3 Pseudo-Serial Schedule Generation Scheme

1: procedure PSSGS
Dg : set of delayed activities
Pj : priority value of activity j
Rj : Fractional part of the priority value Pj
2: g := 0, tg := 0, C := ∅
3: while |J − C| > 0 do
5: Dg = ∅
7: Select j ∈ E(tg ) with lowest priority value Rj
8: if Pj < 1 then
9: Schedule j at tg
10: C := C ∪ {j}
11: recalculate E(tg )
12: E(tg ) := E(tg ) \ Dg
13: else
14: Pj := Pj − 1
15: Dg := Dg ∪ {j}
16: E(tg ) := E(tg ) \ {j}
17: Calculate Ag
18: g := g + 1
19: if |Ag−1 | > 0 then
21: else
22: tg = tg−1
Figure 4.1: Search Space generated by the pseudo-serial schedule generation scheme for a
two activity problem. Each axis corresponds the the priority of one activity. The fractional
parts are priority values, the integer parts are delays.
Figure 4.2: The example problem from Section 3.4.8.8. Five activities using three differ-
ent resource types.
there is no other activity that can be started at that time, using the same resource, activity
E2 must be scheduled at that point by the parallel schedule generation scheme. However
the schedule illustrated in Figure 4.4 is the true optimal solution for the problem. In this
schedule activity E2 is not scheduled at the earliest possible point t1 , but later. It is delayed
beyond E2. This schedule can be generated by the serial schedule generation, giving E1
a lower priority value than E2. It can however also be generated by the pseudo-serial
schedule generation scheme by delaying E2 once, given the priority value of E1 is lower.
This implies that 2 > V (E2) > V (E1) + 1.
If the priority value of E2 would be lower than the one for E1, the delay would yield
Figure 4.3: The only schedule that can be generated by the parallel schedule generation
scheme for the example in Figure 4.2.
Figure 4.4: The optimal solution for the problem in Figure 4.2. Compared to the solution
in Figure 4.3 this is not a non-delay schedule, but requires activity E2 to be delayed once,
which is illustrated by a red double arrow.
the schedule shown in Figure 4.5. This schedule could not be generated by either the
serial or the parallel schedule generation scheme as it is a delayed schedule, but E2 is not
executed at it’s earliest resource and precedence feasible time, given all activities with
lower priority values were already scheduled. If E2 would be delayed a second time, this
would again yield the schedule in Figure 4.4. If however E1 had a lower priority value
than E2, but E2 would be delayed twice, this would result in the schedule illustrated in
Figure 4.6, delaying activity E2 beyond activity F. Both the schedules in Figure 4.5 and
4.6 are neither active nor non-delay schedules and can under no circumstances be optimal.
Since both schedules are contained within the search space, and both schedules do
allow local left shifts of activities, it can be stated, according to the theory presented in
Section 3.4.4.1, that the pseudo-serial schedule generation scheme can generate a superset
of the set of semi-active schedules as defined by Sprecher et al. (1995). It can however not
generate the full set of feasible schedules as the time tg can never be incremented beyond
the finish time of the last activity. Therefore the algorithm can never generate a gap in
Figure 4.5: A solution for the problem in Figure 4.2 that delays activity E2 while E2 has a
lower priority value than E1. This schedule cannot be generated by either of the serial and
parallel schedule generation schemes. It can however be generated by the pseudo-serial
schedule generation scheme.
Figure 4.6: A solution for the problem in Figure 4.2 that delays activity E2 twice, while
E1 has a lower priority value than E2. This schedule cannot be generated by either of
the serial and parallel schedule generation schemes. It can however be generated by the
pseudo-serial schedule generation scheme.
which no activity is active at all. The resulting set of schedules could accordingly be
named the set of pseudo-active schedules. This extends the Lemma defined by Sprecher
et al. (1995) to Lemma 9.
Lemma 9. Let S denote the set of all schedules, SF the set of feasible schedules, SSA the
set of semi-active schedules, SS the set of active schedules, SP A the set of pseudo-active
schedules, and SN D the set of non-delay schedules. Then following holds:
SN D ⊆ SA ⊆ SSA ⊆ SP A ⊆ SF ⊆ S.
Figure 4.7 shows the convergence of the Water Wave Optimization (WWO) algorithm
discussed in Section 4.2 applied on the three discussed schedule generation schemes. The
graph was created using a population of 10 over 1000 generations. As a data set the
50
PSGS
vi =1.1
40 vi =2
vi =3
vi =4
Dev. from opt. in %
30 SSGS
20
10
0
0 200 400 600 800 1000
Iteration
Figure 4.7: Convergence of serial (SSGS), parallel (PSGS) and pseudo-serial schedule
generation schemes on the PSPLIB 30 activity benchmark set. The Water Wave Op-
timization (WWO) algorithm was run using all 3 scheduling schemes. In case of the
pseudo-serial schedule generation scheme, the initial population vi was varied, illustrat-
ing the transition between the serial and parallel schedule generation schemes.
PSPLIB j30 benchmark set was used. The lines show the average deviation in percent
from the known optima of all 480 benchmark problems in the set. It is visible that the
pseudo-serial schedule generation scheme can be used to model the transition between
parallel and serial schedule generation scheme by adjusting the initial population. The
parameter vi defines from which area of the search space the initial population is sam-
pled. A value of 1 implies that the population starts with only non-delay schedules. As
a large part of the optimal solutions in the benchmark set are non-delay schedules, the
parallel schedule generation scheme performs best on average. It is visualized by the grey
dashed line. It can however be stated that for 76 of the 480 problems the solution found
by the pseudo-serial schedule generation scheme was superior. The initial population
was varied between vi = 1.1 and vi = 4. For the first case, the initial population con-
sists of almost non-delay schedules and therefore doesn’t deviate much from the parallel
scheduling scheme’s results. The further the parameter is increased however, the more
the results approach the performance of the serial schedule generation scheme, where the
initial population will always be sampled from the set of active schedules.
Schedule generation schemes generally form vital tools for heuristic algorithms solv-
ing resource constrained project scheduling problems. Compared to the commonly used
serial and parallel schedule generation schemes, the pseudo-serial adaption creates the
transition among both of those schemes, allowing to adjust the characteristics for the
desired purpose. Restricting the search space to the [0, 1] range results in the parallel
schedule generation scheme, while extending it into the positive direction causes it to
converge to the serial schedule generation scheme. It can however be possible to gen-
erate a problem where the number of delays for a certain activity, required to reach the
optimum is so high, that the convergence of the algorithm decreases strongly, because it
extends the search space to a superset of the set of active schedules. Nevertheless does the
algorithm provide a valuable addition to the available schedule generation schemes and
can aid improved results for many heuristic procedures. The runtime complexity of the
pseudo-serial schedule generation scheme is O(N 2 · |R|), hence equal to that of the other
schemes as presented in Section 3.4.4.
4.2 Water Wave Optimization Applied to RCPSPs

A new metaheuristic optimization algorithm named Water Wave Optimization (WWO)
was introduced by Zheng (2015). The initial benchmark of the algorithm involved a
train scheduling problem. The algorithm was then adapted by the author of this thesis to
solve resource constrained project scheduling problems using the pseudo-serial schedule
generation scheme introduced in Section 4.1 since it was already applied to a similar
scheduling problem and delivered promising results.
The algorithm consists of three components, which were inspired by water wave the-
ory, propagation, breaking, and refraction. These components model a number p of punc-
tual waves within the search space, consisting of n dimensions, one for each activity. Each
wave is initialized with a random position within the search space, a height hmax and a
wavelength of λ = 0.5. In each generation of the algorithm, each wave x is propagated,
creating a new wave x0 based on equation 4.1, where rand(−1, 1) samples a number from
the uniform distribution in the range (−1, 1) and λ is the wavelength. L(d) is the length
of the dimension d of the search space.
x0 (d) = x(d) + rand(−1, 1) · λL(d) (4.1)
After each generation the wavelength of each wave x is updated as defined by Equa-
tion 4.2. fmin and fmax are the extreme fitness values of the current population of waves
and f (x) is the fitness value of the current wave x. α is the wavelength reduction coef-
ficient and is a very small number to avoid division-by-zero. This procedure simulates
the movement of waves throughout the search space.
λ0 = λ · α−(f (x)−fmin +)/(fmax −fmin +) (4.2)
If the height of a wave decreases to zero it is refracted based on Equation 4.3. N (µ, σ)
samples a value from the normal distribution with mean µ and standard deviation σ. x∗
is the best solution found so far. This component avoids the search to propagate towards
areas of low fitness.
4.2 Water Wave Optimization Applied to RCPSPs 105
x∗ (d) + x(d) |x∗ (d) − x(d)|

0
x (d) = N , (4.3)
2 2
After refraction the height of the wave is set back to hmax and the wavelength is
updated as defined in Equation 4.4.
f (x)
λ0 = λ (4.4)
f (x0 )
Whenever a wave x reaches a new optimal solution x∗ the wave breaks up into multiple
waves moving in multiple directions. A value k is randomly selected from the range
(k; kmax ), where kmax is a predefined maximum number of breaking waves. At each
randomly selected dimension d it generates solitary waves x0 based on Equation 4.5. If
none of the solitary waves is fitter than x∗ , x∗ remains as the new x0 , otherwise the fittest
of the solitary waves becomes the new x∗ and x0 . This component causes the vicinity of
new extreme points to be evaluated for fitter values.
x0 (d) = x(d) + N (0, 1) · βL(d) (4.5)

The algorithm is listed as pseudo code in Algorithm 4.
Algorithm 4 Water Wave Optimization

1: procedure WWO
2: Randomly initialize a population P of np waves
3: while Stop criterion is not satisfied do
4: for each x ∈ P do
5: Propagate x to a new x0 based on Eq. 4.1
6: if f (x0 ) > f (x) then
7: if f (x0 ) > f (x∗ ) then
8: Break x0 based on Eq. 4.5
9: Update x∗ with x0
10: Replace x with x0
11: else
12: x.h := x.h − 1
13: if x.h=0 then
14: Refract x to x0 based on Eq. 4.3 and 4.4
15: Update wavelengths based on Eq. 4.2
16: return x∗
(Pseudo code from Zheng (2015))
For a more detailed description of the algorithm and its implementation, the reader is
referred to the original paper by Zheng (2015).
For the application of water wave optimization to resource constrained project schedul-
ing problems, the pseudo-serial scheduling scheme, defined in Section 4.1 was used. Ex-
periments showed that the algorithm performs best when the initial population is sampled
from the vicinity of the non-delay region, as shown in Figure 4.1, only. A comparison of
different initial populations is illustrated in Figure 4.7. A parameter surface for the popu-
lation size and the number of generations is illustrated in Figure 4.8. The algorithm was
thoroughly evaluated on the problems of the project scheduling problem library PSPLIB
by Kolisch & Sprecher (1997) and outperforms many other published algorithms.
3
2.5
2
Mean error
1.5
1
0.5
0 200
200 400
400 600
600 800 Iterations
Population size 800
1000 1000
Figure 4.8: Surface of parameter values. Horizontal axes show population size np and
number of iterations i. Vertical axis shows the mean error on the PSPLIB j30 problem
set.
4.3 Iterative Swap-based Limited Depth Search

As described in Section 3.4.8.8 activity swaps can be used to construct search trees based
on an initial schedule. Figure 4.9 shows such an initial schedule consisting of five ac-
tivities using three different resource types. Figure 4.10 and 4.11 show that swapping
activities B and D, or activities A and C, both result in a longer schedule, whereas ap-
plying both activity swaps results in a short schedule, as shown in Figure 4.12. This
illustrates that non-beneficial swaps can be required to arrive at an optimal solution.
4.3 Iterative Swap-based Limited Depth Search 107
Figure 4.9: Example Schedule to illustrate local optimality. Rectangles represent activi-
ties, fill colors represent resources, and arrows represent precedence constraints.
Figure 4.10: The schedule from Figure 4.9 after swapping activities B and D. The swap
increases the makespan of the schedule.
Figure 4.11: The schedule from Figure 4.9 after swapping activities A and C. The swap
increases the makespan of the schedule.
Figure 4.12: The schedule from Figure 4.9 after swapping activities A and C, as well as
B and D. Applying both swaps decreases the makespan of the schedule.
The swap based local search algorithm (Dori, 2016) introduced in Section 3.4.8.8
uses the concept of reasonable swaps and the constraint-based simulation, which is simi-
lar to the parallel schedule generation scheme. For the iterative swap-based limited depth
search algorithm however, the pseudo-serial schedule generation scheme, which was in-
troduced in Section 4.1 is used implicitly. As described in Section 3.4.8.8 the parallel
schedule generation scheme can be modified to collect possible swaps during the sched-
ule generation process. The pseudo-serial schedule generation scheme can additionally
collect possible delays in its process. Algorithm 5 shows the pseudo code for this process.
In contrast to Algorithm 3 in Section 4.1 lines 13-15 and lines 22-24 have been added.
Lines 13-15 check whether, by scheduling one activity, other activities became unfeasible,
hence, whether the set of eligible activities Er (g) reduced in size by more than 1. If that
is the case, each of the newly unfeasible activities can be executed instead of the activity
j, which was just scheduled. Hence all combinations with each of the newly unfeasible
activities x create a possible swap with the scheduled activity (j, x). Additionally lines
22-24 check whether there is more than one active activity in the set of active activities
for the next stage. If that is the case, all activities that have been executed in the current
stage can be delayed. It can however be possible that not all of them can be delayed
simultaneously, as there has to remain at least one active activity at each stage.
Figure 4.13 shows a tree created from the swaps and delays generated by Algorithm 5.
The root node on the top, with index 1, shows the initial schedule from Figure 4.9. Node
2 is generated by swapping activities A and C, while node 4 is generated by swapping
activities B and D. Nodes 3 and 5 are generated by delaying either activity B or C. Node
6 can either be reached by swapping B and D after node 2, or by swapping A and C after
node 4. Hence it would create a duplicate node in the search tree.
In order to be greedy with computational effort it is desired to avoid duplicate nodes
as described by Glover (1990b) in his work on tabu search. It does however require
vast amounts of memory to keep a history of all encountered schedules during the search
process. The C-sequence method and the reverse-elimination method introduced for this
kind of problem in Glover (1990a) however cannot be applied to reliably detect duplicate
schedules, although swaps and delays can be viewed as paired attribute moves. This is
4.3 Iterative Swap-based Limited Depth Search 109
Algorithm 5 Pseudo-Serial Schedule Generation Scheme with Swap Collection

1: procedure PSSGSSC
Dg : set of delayed activities
Us : set of possible swaps
Ud : set of possible delays
Pj : priority value of activity j
Rj : Fractional part of the priority value Pj
2: g := 0, tg := 0, C := ∅
3: while |J − C| > 0 do
5: Dg = ∅
7: Select j ∈ E(tg ) with lowest priority value Rj
8: if Pj < 1 then
9: Schedule j at tg
10: C := C ∪ {j}
11: calculate E 0 (tg )
12: E 0 (tg ) := E 0 (tg ) \ Dg
13: if |E 0 (tg )| < |E(tg )| − 1 then
14: for each x ∈ (E(tg ) \ E 0 (tg ) \ {j}) do
15: Us = Us ∪ {(j, x)}
16: E(tg ) = E 0 (tg )
17: else
18: Pj := Pj − 1
19: Dg := Dg ∪ {j}
20: E(tg ) := E(tg ) \ {j}
21: Calculate Ag
22: if |Ag | > 1 then
23: for each x ∈ (Ag \ Ag−1 ) do
24: Ud = Ud ∪ {x}
25: g := g + 1
26: if |Ag−1 | > 0 then
28: else
29: tg = tg−1
Figure 4.13: Search tree of schedules based on swaps and delays generated using Algo-
rithm 5. The numbers in the upper right corner of each rectangle indicates the node index.
The top schedule is the initial schedule from Figure 4.9. The schedule is changed by ei-
ther a swap or a delay at each edge. Double arrowheads indicate delays. The red rectangle
indicates the optimal solution for the problem. As the depiction contains one undirected
cycle along nodes 1,2,4, and 6, it is not strictly a tree. However, node 6 would practically
only be reached through either node 2 or 4.
caused by the fact that the same schedule can be reached along different paths. It is
possible to reach the same schedule along different paths consisting of entirely different
operations. The schedule in Figure 4.14 can for instance be generated from the schedule
in Figure 4.9 by either delaying activity B, assuming the priority value for D is higher
than that for B, or by swapping activities B and D.
4.4 Evaluation on the Project Scheduling Problem Library (PSPLIB) 111
Figure 4.14: The schedule from Figure 4.9 after either delaying activity B, assuming the
priority value for D is higher than that for B, or after swapping activities B and D.
The idea of the iterative swap-based limited depth tree search algorithm is based on
the ascertainment that the number of swap or delay operations required to escape a local
minimum can be assumed to be low. The number of steps to be taken trying to find a
new optimum is the main parameter of the algorithm. The depth of search per iteration
defines to how many levels the search tree is built up. The best solution found within this
limited depth search tree forms a root node for the next iteration. The algorithm repeats
this process until the root node of the current iteration remains the best known solution.
Compared to search procedures with a fixed tolerance for accepting worse solutions, such
as the tolerance search by Dori (2016), this algorithm can escape even deep local optima,
given it can be escaped with a small enough number of steps. This makes sense because
applying a swap early within a schedule can create an entirely different schedule. Ex-
periments showed that the algorithm compares well to different published algorithms on
the problems of the project scheduling problem library PSPLIB by Kolisch & Sprecher
(1997).
4.4 Evaluation on the Project Scheduling Problem Library

(PSPLIB)
This thesis introduced two new algorithms for the resource constrained project schedul-
ing problem (RCPSP), which were presented in Sections 4.2 and 4.3. For comparing the
efficiency of proposed optimization approaches, the project scheduling problem library
PSPLIB by Kolisch & Sprecher (1996) has become a de facto standard. The PSPLIB
consists of 4 problem sets, j30, j60, j90, and j120 with 30, 60, 90, and 120 activities
per problem respectively. Each of the sets contains 480 problems, except the j120 set,
which contains 600 problems. For each of the j30 problems, the optimal solutions are
known, while for the others, the best known solutions are published. Almost every pub-
lished algorithm for the RCPSP uses the PSPLIB to benchmark the algorithm. Therefore
experiments have been performed on each of the problem sets, ranking the presented al-
gorithms.
The detailed benchmark results are presented in 4 tables, one for each of the PSPLIB
problem sets. The rows labeled RL shows the results of the agent-based reinforcement
learning algorithm by Jedrzejowicz & Ratajczak-Ropel (2014), while the rows labeled
PSO show the particle swarm optimization results published by Zhang et al. (2005a), and
PSO2 refers to the particle swarm optimization algorithm by Chen et al. (2010). The rows
labeled GA list the results of the genetic algorithm presented by Alcaraz & Maroto (2001).
EA is the evolutionary algorithm by Valls et al. (2001) and GH is the greedy heuristic by
Goncharov (2011). HSS is the hybrid scatter search algorithm by Debels et al. (2006).
Rows labeled WWO PS were obtained by the water wave optimization algorithm, which
was presented in Section 4.2, operating on the pseudo-serial scheme, which was presented
in Section 4.1. The rows labeled ILDTS were obtained by using the iterative limited depth
tree search algorithm presented in Section 4.3. For each of the competing algorithms the
best performing parameter set has been selected out of the published results. For details
on the listed parameters, the reader is referred to the original papers. The columns labeled
ropt give the ratio of optimal solutions found for the j30 problem set, while the columns
labeled rbest give the ratio of best known solutions found for the other sets. The x̄ columns
give the mean relative error from the optimal or best known solutions. The columns t give
the mean processing time per solution and the nf columns give the number of fitness
function evaluations or schedule generations performed. Since the processing times were
obtained by the individual authors on different machines, they only yield a rough com-
parison of the run time efficiency, but due to missing nf values for some experiments, the
values were still included in the tables. The t values for the WWO results were measured
on an "Intel(R) Xeon(R) CPU E5-2630 @ 2.30GHz" machine running on a single thread
per experiment. All tables are sorted in increasing order of the mean relative error x̄. For
the WWO rows, np is the population size used and i is the number of iterations of the
algorithm that were performed. All other parameter values were set as recommended in
the paper by Zheng (2015). α = 1.0026, β is linearly decreasing from 0.25 to 0.001, the
maximum number of breaking waves kmax = min(12, na /2) where na is the number of
activities. The maximum wave height hmax = 6. Each table only lists the best results for
each algorithms, an overview of different parameter settings can be found in Appendix
B.1.
Table 4.1 shows the benchmark results for the j30 problem set. The hybrid scatter
search (HSS) algorithm by Debels et al. (2006) performed best on this problem set, but
also the two algorithms developed by the author compare well to other algorithms pub-
lished in literature. While the other algorithms perform better on the data, the run times t
of the agent-based reinforcement learning algorithm by Jedrzejowicz & Ratajczak-Ropel
(2014) are much higher. Same holds for the evolutionary algorithm by Valls et al. (2001).
The genetic algorithm by Alcaraz & Maroto (2001) and the particle swarm optimization
algorithm by Zhang et al. (2005a) converged faster, as they only needed half the num-
ber of fitness function evaluations nf in order to achieve lower error rates. The particle
swarm optimization algorithm by Chen et al. (2010) did not publish the number of fit-
ness function evaluations nf used, but the mean deviation was higher than most tested
configurations of the presented algorithm. When performing i = 1000 iterations on the
Algorithm Parameters ropt x̄ t nf

HSS n = 500000 99.38% 0.01% 7, 160ms 500, 000
RL s = RL123 n/a 0.02% 34, 960ms n/a
WWO PS np = 10000, i = 1000 97.71% 0.04% 156, 677ms 10, 061, 675
EA 93.54% 0.10% 590ms n/a
ILDTS d=3 93.33% 0.14% n/a n/a
GA oc = T P _F BC n/a 0.24% n/a 5, 000
PSO ni = 125 96.40% 0.42% n/a 5, 000
PSO2 n/a 0.54% n/a n/a
Random n = 500000 65.42% 1.16% 49, 967ms 500, 000
Table 4.1: Benchmark results for j30 problem set ordered by the mean deviation x̄ from
the known optimal solutions. The column ropt lists the ratio of optimal solutions found.
nf is the number of fitness function evaluations required per problem. t is the mean
processing time.
Algorithm Parameters rbest x̄ t nf

HSS n = 500000 93.13% 0.07% 19, 610ms 500, 000
EA 86.46% 0.17% 1, 870ms n/a
RL s = RL123 n/a 0.51% 62, 300ms n/a
GA oc = T P _F BC n/a 0.59% n/a 5, 000
WWO PS np = 1000, i = 1000 73.75% 0.87% 69, 396ms 1, 009, 426
ILDTS d=3 64.38% 1.31% n/a n/a
Random n = 500000 57.29% 2.14% 218, 751ms 500, 000
GH 68.33% 2.37% n/a n/a
the best known solutions. The column rbest lists the ratio of best known solutions found.
processing time.
a population size of np = 10000, using the WWO PS algorithm, the error is x̄ = 0.04%,
while in 97.71% of the problems the global optimum was found. This outperforms the
results of the PSO, EA, and GA algorithms. For comparison with the PSO and GA algo-
rithms two experiments with each nf ∼ 5000 schedules generated have been conducted
yielding a minimum mean error of x̄ = 1.23% which is slightly worse than the compared
algorithms. Generally it was observed that more than i = 1000 iterations only have a
small effect on the outcome, while the population size plays a more important role. This
was illustrated by a performance surface in Figure 4.8 in Section 4.2, which shows the
mean error over the space of the parameters population size and iterations on the j30
problem set. It shows that a certain number of iterations is required, after which a larger
population yields better improvements. The ILDTS algorithm performs slightly worse
than the WWO PS algorithm, finding 93.33% of the global optima with a mean error of

HSS n = 500000 91.04% 0.07% 38, 870ms 500, 000
EA 88.75% 0.16% 4, 900ms n/a
RL s = RL123 n/a 0.98% 77, 290ms n/a
WWO PS np = 1000, i = 1000 71.67% 1.61% 133, 577ms 1, 011, 275
ILDTS d=3 65.42% 1.62% n/a n/a
Random n = 500000 57.71% 2.64% 353, 538ms 500, 000
processing time.

HSS n = 500000 72.33% 0.28% 69, 570ms 500, 000
EA 50.17% 0.75% 35, 000ms n/a
GA oc = T P _F BC n/a 1.36% n/a 5, 000
RL s = RL123 n/a 2.91% 202, 410ms n/a
WWO PS np = 1000, i = 1000 25.33% 5.48% 259, 471ms 1, 020, 095
ILDTS d=2 11.83% 6.18% n/a n/a
Random n = 500000 14.17% 7.41% 482, 920ms 500, 000
processing time.
x̄ = 0.14% using a search depth of d = 3.

Table 4.2 lists the results of running the different algorithms on the j60 problem set
of the PSPLIB. In its best configuration with a population of np = 1000 waves and i =
1000 generations, the WWO PS algorithm finds 73.75% of the best known solutions
and deviates by 0.87% on average. While outperforming the published results for the
greedy heuristic by Goncharov (2011) and the particle optimization algorithm by Chen
et al. (2010), the other algorithms perform slightly better on the benchmark data. Due
to missing data on the number of fitness function evaluations nf a fair comparison of
the evolutionary and reinforcement algorithms was not possible. The presented WWO
PS algorithm can be configured to run much faster than the other algorithms, while the
drawback on the quality results is very small. The ILDTS algorithm, given a depth of
d = 3 found 64.38% of the best known solutions with an average deviation of 1.31% and
outperforms the GH and PSO2 algorithms.
For the j90 benchmark set only results for four of the other algorithms were published,
which are compared in Table 4.3. With a ratio of 71.67% of the best known solutions
found and a mean deviation of 1.61% the WWO PS algorithm delivers good results which
outperform the particle swarm optimization algorithm by Chen et al. (2010), but could
neither match the reinforcement learning algorithm by Jedrzejowicz & Ratajczak-Ropel
(2014) nor the evolutionary algorithm by Valls et al. (2001) in quality of the results. With
little drawback though, the presented algorithm can be configured to run much faster
than the other algorithms. Unfortunately the number of fitness function evaluations nf
required to obtain the results was not published for either of the other algorithms. The
ILDTS managed to find 65.42% of the best known solutions and deviated by 1.62% on
average.
Table 4.4 compares the results of the presented algorithms to the other algorithms on
the most complex j120 problem set containing 600 problems with 120 activities each.
While the genetic algorithm (GA) and the hybrid scatter search (HSS) gave outstanding
results, the other algorithms published very high run times for their results. In the best
configuration that was tested, np = 1000, i = 1000 the WWO PS algorithm was only
able to outperform the particle swarm optimization algorithm by Chen et al. (2010), but
with a mean error of x̄ = 5.48% still produced very good results. The ILDTS algorithm
falls back, with only 11.83% of the best known solutions found, and an avarage deviation
of 7.41%, given a search depth of d = 2. Experiments with larger search depths were not
feasible due to the immense number of possible swaps within the schedules.
While outperformed by some of the algorithms in literature, especially the hybrid
scatter search algorithm by Debels et al. (2006), the presented algorithms provide a valu-
able addition to the available arsenal of RCPSP heuristics. Especially the pseudo-serial
schedule generation scheme offers the possibility to be used with a wide variety of other
metaheuristic algorithms.
4.5 Uncertainty Extension for the Project Scheduling

Problem Library PSPLIB
Construction processes rarely have a constant duration that is known a-priori. Instead
their durations are estimated based on experience and previously collected sample data.
From this sample data probability distributions can be estimated. The Project Schedul-
ing Problem Library PSPLIB by Kolisch & Sprecher (1997) only contains problems with
fixed durations. In order to determine the practical capabilities of an developed algorithm
however, its performance on problems involving uncertainty is of high interest. Bershtein
et al. (2015) extended some of the PSPLIB problems by using a 6-point piecewise linear
membership function as defined by Rommelfanger (1990). They did however not pub-
lish the used distributions, such that they can be used for comparison studies. Another
approach by Fu et al. (2012) added normal distributed delays with mean 0 and standard
deviation 1 to each activity of the PSPLIB problems.
However, as described in Section 2.2.1, the beta distribution is the most suitable rep-
resentation of uncertain durations in construction operations. Therefore beta distribution
parameters were created for each problem in the PSPLIB. For each activity of each prob-
lem in the single mode sets j30, j60, j90, and j120 a maximum delay dˆj , and the αj
and βj parameters for the beta distribution have been pseudo-randomly created5 . The ex-
tended problem sets are named j30u, j60u, j90u, and j120u respectively. The delays
are uniformly sampled integers from the interval (1, 5), and the parameters α and β are
uniformly sampled integers from the interval (1, 10). In order to get a sample for the
duration of activity j d¯j , Equation 4.6 can be used, where B(αj , βj ) takes a sample from
the beta distribution with parameters αj and βj . Since many implementations are based
on integer durations, the sampled duratios are rounded to integers. This has the advantage
of the delays having strict bounds which creates the advantage that the known optima for
each of the PSPLIB’s problems form a lower bound for each of the uncertainty adaptions,
which does not hold for the normal distributed approach by Fu et al. (2012).
d¯j = dj + [dˆj · B(αj , βj )] (4.6)
As an example, each activity’s duration was plotted for the j30u problem with instance
1 and parameter 1 in Figure 4.15. The resulting possible schedules are plotted in Figure
4.16. The figure shows the Monte-Carlo results for 100, 000 runs. Each plot shows the
likelihood of an activity being active at the respective time. The precedence constraints
of the activities are illustrated in Figure 4.17. Figure 4.18 shows the distribution of the
overall makespan over 1, 000, 000 runs.
In order to compare different algorithms on this extended benchmark set, a number of
eight objective functions have been designed, which are listed in Table 4.5. Each objective
is based on performing a certain number of Monte Carlo runs of a resulting schedule. O1
defines the first objective function of the minimum makespan of all the K Monte Carlo
runs. O2 defines the second objective as the mean makespan, while O3 defines the worst
5
The generated values have been published in the git repository hosted at https://github.com/tumcms/
upsplib to be used for comparison studies.
encountered makespan. O4 defines the standard deviation of the set of the Monte-Carlo
makespans. O5 defines the objective as the sum of the mean makespan and the standard
deviation. O6 defines the objective as the mean makespan added to twice the standard
deviation. O7 is the worst encountered makespan added to the standard deviation, while
O8 adds twice the standard deviation. O5 till O8 are multi-criteria objectives using a
weighted sum as presented in Section 3.2.2.4.
X mk
O1 = min(mk ) O2 =
k∈K
k∈K
|K|
s
X m2 X mk
k
O3 = max(mk ) O4 = −( )2
k∈K
k∈K
|K| k∈K
|K|
O5 = O2 + O4 O6 = O2 + 2 · O 4
O7 = O3 + O4 O8 = O3 + 2 · O 4
Table 4.5: The eight objective functions of the UPSPLIB problem library.
4 3 2
0 0 0
0 7 14 0 7 14 0 7 14
3 3 2
0 0 0
0 7 14 0 7 14 0 7 14
2 3 3
0 0 0
0 7 14 0 7 14 0 7 14
3 2 2
0 0 0
0 7 14 0 7 14 0 7 14
3 4
2
0 0 0
0 7 14 0 7 14 0 7 14
2 3 3
0 0 0
0 7 14 0 7 14 0 7 14
3
1 2
0 0 0
0 7 14 0 7 14 0 7 14
2 3 3
0 0 0
0 7 14 0 7 14 0 7 14
3 3
2
0 0 0
0 7 14 0 7 14 0 7 14
3 2 1
0 0 0
0 7 14 0 7 14 0 7 14
Figure 4.15: Uncertain durations of the 30 individual activities for the j30u problem with
instance 1 and parameter 1. Each activity is delayed using a beta distributed random vari-
able. The horizontal axes show the duration, the vertical axes the probability. Activities
are arranged from left to right, then top to bottom.
1 1 1
0 0 0
0 20 40 60 80 0 20 40 60 80 0 20 40 60 80
1 1 1
0 0 0
0 20 40 60 80 0 20 40 60 80 0 20 40 60 80
1 1 1
0 0 0
0 20 40 60 80 0 20 40 60 80 0 20 40 60 80
1 1 1
0 0 0
0 20 40 60 80 0 20 40 60 80 0 20 40 60 80
1 1 1
0 0 0
0 20 40 60 80 0 20 40 60 80 0 20 40 60 80
1 1 1
0 0 0
0 20 40 60 80 0 20 40 60 80 0 20 40 60 80
1 1 1
0 0 0
0 20 40 60 80 0 20 40 60 80 0 20 40 60 80
1 1 1
0 0 0
0 20 40 60 80 0 20 40 60 80 0 20 40 60 80
1 1 1
0 0 0
0 20 40 60 80 0 20 40 60 80 0 20 40 60 80
1 1 1
0 0 0
0 20 40 60 80 0 20 40 60 80 0 20 40 60 80
Figure 4.16: Schedule for the j30u problem with instance 1 and parameter 1, as shown in
Figure 4.15, using the default activity order with the parallel schedule generation scheme.
Each plot corresponds to one activity and shows the probability of an activity being active
at a certain time. The plots are based on 100, 000 runs. Activities are arranged from left
to right, then top to bottom.
1 (0)
3 (4)
8 (9) 7 (5) 4 (6)
12 (2) 27 (8) 9 (2) 10 (7)
14 (3) 13 (6) 16 (10) 5 (3) 2 (8)
17 (6) 21 (2) 18 (5) 11 (9)
28 (3) 22 (7) 20 (7) 26 (7) 15 (9)
19 (3) 31 (2) 23 (2) 25 (3)
24 (3) 6 (8)
29 (7) 30 (2)
32 (0)
Figure 4.17: Dependency graph of the 30 activities for the j30u problem with instance
1 and parameter 1. The numbers in parentheses are the minimum durations. Activities 1
and 32 are added source and sink activities with duration 0.
150000
Number of occurrences
100000
50000
0
0 20 40 60 80
Duration
Figure 4.18: Uncertain duration of a schedule for the j30u problem with instance 1 and
parameter 1, as shown in Figure 4.15, using the default activity order with the parallel
schedule generation scheme. The vertical axis shows the number of occurrences out of
1, 000, 000 runs.
This Section contains three tables for the j30u problem set. Each table has one column
for each of the objective functions defined previously. Furthermore there are only four
rows for the first four objective functions, as the other functions are combinations of the
first four. The columns correspond to the optimization objectives that were used, and the
rows correspond to the output values. Each table was populated using simulation results
of the best solutions found, according to the respective objective functions, using the water
wave optimization algorithm with a population of p = 10 waves and n = 1000 iterations.
For each objective function evaluation, 10 Monte Carlo runs were carried out. The final
solution for each of the optimization runs was evaluated 10, 000 times in a Monte Carlo
way.
Table 4.6 shows the minimal deviations from the optimal solutions of the correspond-
ing j30 PSPLIB problem for the first three rows, while the last row shows the standard
deviation. The first row shows the minimum deviation of the best simulation runs for
each solution. The second row shows the mean deviation and the third row shows the
maximum deviation. Table 4.7 shows the same results, but the mean values of the 10, 000
runs, while Table 4.8 shows the worst solutions encountered during the Monte Carlo runs.
Looking at Table 4.7 showing the mean results several things can be noticed. The low-
est minimum makespan is achieved when using the minimum makespan O1 or the mean
makespan O2 as objective function. The lowest maximum makespan is achieved when us-
ing the maximum makespan O3 as an objective function. But similar results are achieved
when minimizing the sum of the mean makespan and the standard deviation (O5 ) or max-
imum makespan and the standard deviation (O7 ). Using the standard deviation O4 as an
objective function generally returned poor results. This can be traced back to the stan-
dard deviation creating a very discontinuous search space. When looking at the worst
case Table 4.8 is is visible that the objective function that minimizes both the worst case
makespan, as well as the worst case standard deviation is the sum of the mean makespan
and the standard deviation (O5 ). Additionally this objective function also acchieves the
lowest worst case and mean best case values in Table 4.6. More detailed results are pre-
sented in Appendix B.2 and the precise solutions have been published to github6 .
6
https://github.com/tumcms/upsplib
XXX
X Objective
XXX O1 O2 O3 O4 O5 O6 O7 O8
Result XXX
XX
O1 (min) 1.05 1.02 1.04 1.04 1.03 1.03 1.04 1.04
O2 (mean) 1.11 1.11 1.11 1.13 1.11 1.11 1.12 1.11
O3 (max) 1.16 1.16 1.16 1.21 1.16 1.18 1.20 1.18
O4 (stDev) 0.89 0.89 1.02 1.03 0.96 0.98 0.95 0.92
Table 4.6: Best solutions of water wave optimization on the uncertainty-extended PSPLIB
j30u set for the different objective functions O1 to O8 . The result measures have been
obtained through 10,000 Monte Carlo runs of the final solution.
XXX
XXX Objective
Result XXX
O1 (min) 1.21 1.21 1.22 1.54 1.22 1.22 1.22 1.22
O2 (mean) 1.33 1.32 1.32 1.68 1.32 1.32 1.32 1.32
O3 (max) 1.49 1.47 1.46 1.85 1.46 1.47 1.46 1.47
O4 (stDev) 2.82 2.41 2.09 2.47 2.11 2.17 2.09 2.09
Table 4.7: Mean results of water wave optimization on the uncertainty-extended PSPLIB
XX
XXX Objective
Result XXX
X
O1 (min) 1.46 1.46 1.49 2.36 1.48 1.53 1.56 1.49
O2 (mean) 1.68 1.59 1.60 2.62 1.62 1.67 1.66 1.62
O3 (max) 1.91 1.97 2.11 2.77 1.88 2.09 2.14 2.06
O4 (stDev) 12.79 9.44 7.04 8.30 5.72 9.87 8.01 6.24
Table 4.8: Worst solutions of water wave optimization on the uncertainty-extended

PSPLIB j30u set for the different objective functions O1 to O8 . The result measures
have been obtained through 10,000 Monte Carlo runs of the final solution.
This Section illustrated that the uncertainty extension of the PSPLIB problem library
enriches the commonly used benchmark problems by adding uncertain delays to each
problem and therefore allowing to evaluate an algorithms performance for RCPSPs with
uncertain activity durations. The added objective functions allow the heuristic optimiza-
tion of different statistical characteristics of the problems. The inclusion of the schedule
durations’ standard deviation generates a more chaotic search space, creating an added
challenge for heuristic optimization procedures.
4.6 Data Fusion and Reactive Scheduling

Once changes in the circumstances of a construction project have been detected, or de-
lays have occurred, a replanning process might be necessary in order to account for the
changes. The critical path of the schedule and the float times of the activities, as de-
scribed in Section 3.4.6 play an important role in that process. If a process is delayed
beyond its float time, or the process is critical, a replanning process has to occur. Any
activity that has already finished can be neglected in that process. Same holds for any
precedence constraint linked to the removed activities. If any of the removed activities
did take an unanticipated amount of time, the performance factors, and possibly the prob-
ability distributions, of the involved resources might need to be updated, given there are
future processes that are affected by them. Activities that are currently in progress need
to be shortened by the amount of time the process has been in execution. If the percentile
progress of an activity that is currently in execution is known, a performance factor up-
date might change the duration and probability distributions of that activity. Once all
those changes have been incorporated into the project model, a new optimization run can
be performed, yielding an updated schedule, taking the newly arisen circumstances into
account.
4.7 Summary
This Chapter introduced a new schedule generation scheme, the pseudo-serial schedule
generation scheme, that combines advantages of both the commonly used schemes, the
serial and parallel schedule generation schemes. It has been shown that the new scheme
can generate all active schedules, as does the serial scheme, but is computationally more
efficient and can be used to collect possible swaps and delays during schedule generation,
as does the parallel schedule generation scheme. Furthermore this Chapter introduced two
novel scheduling algorithms. The Water Wave Optimization (WWO) algorithm by Zheng
(2015) was applied to RCPSPs and the iterative swap-based limited depth search has been
developed. Both methods have been shown to compete with other algorithms from litera-
ture, however the WWO algorithm proved to be much stronger and computationally more
efficient in finding good solutions to the problems of the PSPLIB. Finally an extension
of the PSPLIB has been developed, adding uncertainty using beta-distributed delays to
each activity in each problem. All values have been published to allow comparisons to be
performed by other researchers. The extension allows scheduling algorithms to be eval-
uated on problems with uncertain performance factors and activity durations. It has been
shown that the presented algorithms can handle the extended problems and therefore can
be applied to real-life problems involving data acquisition with uncertainty in the context
of a reactive scheduling system.
125
5 State-of-the-Art in Data
Acquisition on Construction Sites
"Not everything that counts can be counted and not everything that can be
counted counts." - Cameron (1963)
Large construction projects are built over long periods of time, sometimes years pass
before the project is finished. As such projects consist of many individual activities it can
be difficult to maintain an overview over the state of all of them. Additionally construc-
tion projects are commonly subject to unforeseen events that affect the building process.
Therefore copious research was dedicated to data acquisition techniques for construction
activities. For the reactive scheduling approach, data feedback is a vital ingredient. The
acquired progress information can be used to update the model of the construction project
by updating performance factors and activity durations. This data is important to keep the
model of the construction project synchronized and therefore allow for reactive planning
and prognosis.
5.1 Protocol-based Tracking

Nowadays it is still common practice to manually log construction progress on paper
or in spreadsheats. Kagioglou et al. (2000) formalized the generic design and construc-
tion process protocol (GDCPP) which is a framework model, describing processes on
a construction site in a formal manner, such that each involved party’s interests are in-
cluded. Furthermore it facilitates computer-driven processing and optimization of the
project structure. Such a formal process description can also be used to precisely log the
progress of each involved activity. While in many cases the use of manual logging of
progress information is common, Saidi et al. (2002) discusses the value of using hand-
held devices such as tablets in order to aid with this process. This process was later
improved by special purpose android applications such as "BIM Track", which was pre-
sented by Marzouk & Zaher (2015). This application allows the end user to input status
and progress information, such as dates, completion ratio, and actual cost, by using a
building information model. Generally though, human-driven approaches can be error-
prone and cause costs by requiring manpower on site.
126 5 State-of-the-Art in Data Acquisition on Construction Sites
5.2 Machine data

Many machines nowadays are equipped with sensors to evaluate their performance. For
instance, modern drilling rigs record the drill depths, pressure levels, and torque at certain
time intervals (Obergrießer et al., 2009). The data recorded by those sensors can be used to
collect information about the activities that were performed using that machine (Kargul
et al., 2015). A patent for such a system was filed by Likins et al. (1999). It uses the
sensors installed within a drilling rig to record data on the performance of pile installation
systems. Based on the retrieved values, such as drill depth, speed, pressures and velocity
the individual piles, that were produced, can be distinguished and linked to respective time
windows. Viljamaa & Peltomaa (2014) describes a more holistic construction progress
control system also taking excavator and truck mass realization data into account. The
use of machine data has a fairly limited focus as it is only useful when machines are
deeply involved into the activities to be observed and the sensor data can be used to derive
information on the activity performed as well as its status. For instance the rotation and
load of a tower crane does not deliver sufficient information to infer what kind of activity
it is lifting material for.
5.3 Laser Scanning

A commonly used method to acquire three dimensional data of construction sites is laser
scanning. Cheok & Stone (1999) and Cheok et al. (2000) proposed the use of laser scan-
ning in order obtain measurements from construction sites without interfering with the
ongoing activities. They compared the measurements with a 3D model of the project to
notice deviations. Kern (2002) proposed the use of laser scanner technology to calcu-
late precise volume information from the resulting 3D point clouds of earthwork oper-
ations. Hashash et al. (2005) and Su et al. (2006) advanced the approach towards ob-
taining geotechnical measurements during urban excavation operations. This approach
allows to obtain progress information of the excavation activities. Shih (2003) and Shih
& Wang (2004) used laser scanners to compare as-built with as-planned information based
on the resulting point clouds. This allows the tracking of objects on site and to determine
whether a part was already placed at its destination position. Tang et al. (2010) reviews
methods for detecting objects in building information models (BIM) within laser-scanned
point clouds based on different geometrical models. Those methods were successfully
applied for dimensional compliance control by Bosché (2010). In contrary Dore & Mur-
phy (2014) present an approach to create building models from laser scans of façades.
The wide range of possible uses of laser scanned point clouds make this technology an
attractive choice. The laser scanner however is expensive, needs trained personnel, and
multiple scans may be required to capture all relevant locations. Through stitching of the
resulting point clouds, small errors can be created which make it troublesome to locate
small objects on site (Matiukas & Miniotas, 2011).
5.4 Photogrammetry 127
5.4 Photogrammetry
Ullman (1979) and Webb & Aggarwal (1982) developed an approach, called structure-
from-motion (SFM), that allows to create three dimensional point clouds from photographs.
Later this idea was developed further into a system called VisualSFM by Wu (2011). In
combination with the patch-based multiview stereo (PMVS) algorithm published by Fu-
rukawa & Ponce (2010) this method allows to generate very high quality colored point
clouds from unordered photographs of an object.
In order to generate such point clouds, a method for finding common features in im-
ages is required. For this purpose the Scale Invariant Feature Transform (SIFT) (Lowe,
1999) and the Speeded Up Robust Features (SURF) (Bay et al., 2006) algorithms are
most commonly used. Both algorithms create descriptions of local geometric features
within images that are invariant to scale and rotation. Furthermore they are robust against
changes in illumination, noise and slight deformation. Each feature can be compared to
other features in order to identify similar points in different images. Figure 5.1 a) and b)
shows two illustrative images of a mountain site. Figure 5.1 c) and d) shows how features
in the images can be mapped to each other. The blue dashed lines show the identified
features. This technique is commonly used for panorama stitching, but also finds use in
the SFM approach.
Figure 5.1: Matching of image features. a) and b) show two input images. The blue
dashed lines connect matching features in c) and d).
The phrase structure-from-motion implies that 3D information can be extracted from

2D images by motion of the camera. More precisely, many such images have to be taken
and their features have to be matched among them. Figure 5.2 illustrates that a feature
located in a single 2D image can not generally be located in 3D space (See Figure 5.2).
In order to make use of the motion of the camera in between multiple shots it is required
to locate the same feature points within the images. In theory this requires the features
2
of every image to be compared to the features of every other image, which requires n2
image comparisons for n images. For large objects where many images are required, this
can take substantial amounts of time. In order to reduce this effort, Silpa-Anan & Hartley
(2008) have developed an optimized matching algorithm based on the k-d tree algorithm
by Beis & Lowe (1997) and Lowe (2004).
Figure 5.2: The location of a feature in a single two dimensional image does not reveal
its location in three dimensional space.
If the same feature has been located in two different images while the positions where
the images were taken, and the parameters, as well as the orientation of the camera were
known, it is possible to locate the feature in 3D space. This is illustrated in Figure 5.3.
The problem becomes more complex if the camera positions are unknown. Depending
on whether the camera parameters are known and whether it is known that the same
camera setup was used for all recorded images, an initial image pair needs between six
and eight feature correspondences and each added feature requires between two and three
correspondences (Astrom et al., 1999; Hartley & Zisserman, 2002). Figure 5.4 illustrates
two feature correspondences in three images.
Small deviations in the detection of features and the quality of the recorded images
cause the features to not match up precisely. For this purpose the bundle adjustment algo-
rithm is used to find the optimal geometry minimizing those errors among the analyzed
images (Brown, 1976; Triggs et al., 2000). Repeating this procedure for several images
results in a point cloud of the recorded scenery.
Finally the density of the resulting point cloud can be increased by using dense stereo
reconstruction methods such as the patch-based multi-view stereo algorithm by Furukawa
& Ponce (2010). For an overview of such methods the reader is referred to the paper by
Ahmadabadian et al. (2013).
Tuttas et al. (2014) used photogrammetry to monitor the progress of construction
projects using building information models (BIM). The work was later extended by using
geometric constraints which were derived from BIMs in order to fill in missing informa-
5.4 Photogrammetry 129
Figure 5.3: The locations of a feature in two two dimensional images reveal its true loca-
tion, given the locations and parameters of both cameras are known.
Figure 5.4: Example for two features in three images.
tion (Braun et al., 2015a,b; Stilla et al., 2015; Tuttas et al., 2015).
A method for measuring excavated volume from photogrammetric point clouds of
excavation pits has been developed by the author and is presented in Section 6.1.2.
5.5 Barcodes and QR-Codes

Another common approach for material tracking on construction site is the placement of
QR or barcodes on parts. The personnel, placing the part, then scans the codes to mark the
respective activities as done. Such an approach was first proposed by Echeverry (1996),
later refined and combined with geographic information systems (GIS) by Cheng & Chen
(2002) and are still subject to ongoing research. (Lin et al., 2014; Wang et al., 2014). The
approach only allows the tracking of parts that are preproduced and placed on site during
the building process. Parts that are produced on site, such as concrete parts can not be
monitored using this approach. Furthermore a code can be damaged during placement of
the part, or be painted over in the following building process. Furthermore the process of
placing and scanning the codes can hardly be automated.
5.6 Radio-based tracking

The use of radio-based tracking techniques on the other hand is more feasible for automa-
tion. In practice three different types of radio based tracking techniques are used.
5.6.1 GPS
The global positioning system (GPS) is based on tracking devices which can be located
with a few meters of deviation. In 1995 when the GPS system was officially activated,
one of the first commercial use cases was the tracking of heavy agricultural machines
(O’Connor et al., 1995). This approach was first presented in a construction site envi-
ronment by Naresh & Jahren (1997), where trucks were equipped with GPS trackers. At
this point equipping a single truck with this technology created costs of several thousand
dollars and was therefore hardly feasible on a large scale. When the technology became
more widespread and costs dropped drastically, its use became more realistic and new
approaches were envisioned. Oloufa et al. (2003) presented a method to not only track
trucks, but also on-site equipment and additionally showed that the method can be used
to prevent collisions. Furthermore Li et al. (2005) showed that the technology can reduce
waste, improve efficiency, and reduce costs as a result. The idea proceeded to also cap-
ture the positions of materials on the site, as presented by Song et al. (2005). Pradhananga
& Teizer (2013) presented methods for detailed spatial-temporal analysis of the obtained
data, to improve efficiency of the on-site operations even further. All those technologies
are widely used in many areas today. However, since the costs of GPS trackers still aren’t
negligible, the technology still isn’t efficient for material tracking, especially for small
parts.
5.6.2 RFID
A much cheaper approach to tracking materials on-site is radio frequency identification

(RFID), which is based on passive trackers which are very cheap to produce and have a
5.7 Visual Tracking 131
tiny form factor. If they are activated by surrounding radio waves, they broadcast their
hardcoded ID which can then be received by a reader device. This allows to register RFID
tags in the area surrounding a reader. Jaselskis et al. (1995) presented different possible
applications of RFID on construction sites. Similar to GPS, a truck can be equipped with
a RFID tag, but in comparison it cannot be continuously tracked with this technology,
but the times it passes certain points, referred to as gates, can be logged. This would for
instance be the times it enters and leaves a construction site. Furthermore RFID tags can
be cast into prefabricated concrete parts, whose entrance onto the site can then be logged.
Moreover all tagged and placed parts on the site can easily be verified by a walk-through
using a RFID reader device, logging all surrounding tags (Song et al., 2005). Costin
et al. (2012) developed this technology further into a system that automatically reports the
progress of ongoing activities based on the location of RFID tags placed on materials and
personnel. Kiziltas et al. (2008) analyzed the practical implications of this technology.
They also discuss the active variant of RFID, which is commonly implemented using
ultra-wideband technology. Klaubert & Günthner (2011) presented a holistic RFID-based
approach for construction information and communication systems.
5.6.3 Ultra-Wideband Radio
Compared to passive RFID, ultra-wideband (UWB) technology is active. The UWB de-
vice sends radio waves to allow a receiver infrastructure to track its location. Compared to
passive RFID, UWB allows precise location tracking in three dimensions. Furthermore,
it is more precise than GPS and works in occluded areas, such as indoor spaces (Teizer
et al., 2007b). As materials are often stored in indoor storage places, GPS tracking is not
possible in those cases. The system has been used for performance measurements (Cheng
et al., 2011; Saidi et al., 2011), resource tracking (Teizer et al., 2007a), and personnel
tracking (Sahinoglu, 2008).
5.7 Visual Tracking

Compared to the previously explained methods for tracking objects on construction sites,
the use of visual information is perhaps the most intuitive method. It is however much
more complicated to be automated.
5.7.1 Range Imaging
One approach to visual tracking, which is similar to the previously explained method of
3D scanning, is range imaging. A range image is a two dimensional image, that is en-
hanced by adding depth information to each pixel. There are different methods to generate
this depth information. The best known system is the use of stereo triangulation, which is
also used by the human vision apparatus and accordingly for recording 3D movies. An-
other common technique is the use of structured light, projecting a special light pattern
onto the scene which can be used to retrieve depth information. The most advanced meth-
ods use a coded aperture (Levin et al., 2007) or so-called time-of-flight (ToF) methods
(Oggier et al., 2004). Time-of-flight methods create a single pulse of light and measure
the time it takes to be reflected from the different surfaces in the scene. As ToF works in
real-time and is very precise it can be used for real-time tracking of resources on site.
Teizer et al. (2007a) presented the use of range imaging for maintaining a occupancy
grid of a construction site, while Kahlmann et al. (2007) presented its use for personnel
tracking. The idea was extended towards the creation of as-built building models by
Bosche & Haas (2008). Teizer (2008) and Gonsalves & Teizer (2009) proposed the use
for safety analysis and human motion analysis. An approach for 3D object detection for
heavy equipment tracking was introduced by Son et al. (2010). Ng et al. (2005) describe a
new technology, the plenoptic camera, which records images or video in three dimensions
using a single camera. As the resolution of such cameras is still limited and the costs are
very high, they were not yet subject to research in the area of construction tracking.
Generally the use of range imaging techniques is cost intensive and creates large
amounts of data that are hard to process automatically. Therefore two-dimensional imag-
ing techniques are still in the focus of researchers.
5.7.2 Video Analysis
Many processes on construction sites can be monitored visually. A human viewing the
site all day from a vantage point, would be able to monitor most processes. Especially
in the shoring and excavation phase, all involved personnel and machines are mostly vis-
ible. Therefore computer vision-based techniques are of great interest for monitoring
those processes automatically. The use of computer vision techniques on video record-
ings has a long history of research. In the seventies, when computers first became capable
of processing video sequences Flachs et al. (1976) and Flachs et al. (1977) presented an
automatic object tracking system that formed the basis of decades of following research.
Later they managed to run the tracking system in real-time (Gilbert et al., 1980). Despite
those early advances in video tracking technology, there are still many challenges subject
to ongoing research. Song et al. (2006) published an approach to video based tracking on
construction sites to both measure and improve the performance of construction opera-
tions. Teizer & Vela (2009), Yang et al. (2009) and Yang et al. (2010) present an approach
for automated video-based personnel tracking. A possible approach to assess the pro-
ductivity of construction operations is to identify the status of different activities by the
postures of involved workers (Ray & Teizer, 2012). Furthermore the tracking of resources
and materials can be subject of video monitoring approaches (Gong & Caldas, 2010). On
many construction sites tower cranes play a key role. Yang et al. (2011) presented an
approach for vision-based crane tracking for the analysis of construction activities. An-
other common key activity of construction projects are earthwork processes. Ogunmakin
et al. (2013) presented a method using video cameras to quantify interactions between
excavators and dump trucks. A holistic system, including video based monitoring of the
construction activities was presented by Leung et al. (2008).
Section 6.2 presents a novel method, that is based on video-based tracking, as well as
photogrammetry.
5.8 Summary 133
5.8 Summary
This Chapter presented an overview of the state-of-the-art of progress tracking technolo-
gies that can be used on construction sites. While manual protocol-based tracking is still
widely used in practice, it is also possible to automatically record machine data and ex-
tract information about the executed activities from that data. It has been shown that this
information can be used to calculate performance factors for the involved resources. Fur-
thermore laser scanning and photogrammetry have been proven effective in generating
point clouds with sufficient quality to detect construction elements and measure the as-
built performance. Additionally video-based approaches were presented, and have been
shown to be effective in tracking construction assets. The presented methods have not
been used to monitor excavation processes by obtaining absolute measures. While there
are methods to count the number of dump trucks entering and leaving a site, the exact
volume of excavated soil is a more precise measure. The following Chapter introduces
improved vision-based methods for this purpose.
134 6 Improved Vision-based Progress Monitoring of Excavation Processes
6 Improved Vision-based Progress

Monitoring of Excavation
Processes
"There’s nothing like matter of fact; seeing is believing." - Arbuthnot (1712)
Many construction projects strongly depend on shoring and excavation processes.

Main actors in those processes are excavators, dumpers, and shoring machines. There
are several approaches in literature using radio frequency based tracking of machines and
trucks, such as RFID or GPS (Cheng et al., 2011; El-Omari & Moselhi, 2011; Klaubert
et al., 2010; Lu et al., 2007; Oloufa et al., 2003; Saidi et al., 2011). However, in practi-
cal excavation projects the dump trucks are often hired on demand on a daily basis and
are therefore difficult to equip with a radio chip. Furthermore counting the number of
dump trucks or the movement of machines does not yield detailed progress data for the
involved processes. This Chapter introduces two vision-based techniques for this purpose,
which can be used to obtain detailed progress information about excavation processes and
determine underlying problems when delays occur. The resulting data can be used to up-
date performance factors for the tracked processes as described in Section 2.1, as well as
other similar processes that need to be performed in the future. Furthermore, the knowl-
edge about the source of the delay allows the decision makers to put countermeasures
into place. Additionally it allows feedback of real-time data into a construction project
model and therefore facilitates updated prognoses and reactive optimization of construc-
tion schedules.
6.1 Excavation Tracking using Point Clouds

The most vital measure for the progress of an excavation process is the amount of exca-
vated soil, which can be measured by the volume of the excavated pit. As the excavated
volume is surrounded by a surface of soil, it can be described by a set of points lying on
that surface. When multiple such sets are known for different time instances, the volume
of the soil excavated within the time period between the samples can be determined. The
two basic methods for generating such point sets, or point clouds, are described in the
following two Sections, followed by methods to extract the desired metrics from the gen-
6.1 Excavation Tracking using Point Clouds 135
erated point clouds. As a result up-to-date performance factors for the involved processes
can be calculated.
6.1.1 Laser Scanning
As described in Section 5.3 laser scanning can be used to generate three dimensional
point clouds. The laser scanner is placed near the excavation pit to record a laser scan.
Due to the concave shape of an excavation pit, a single laser scan can hardly capture
the full surface of the pit, as is illustrated in Figure 6.1. Additionally machines, such as
excavators and dumpers commonly operate within the pit and create occluded areas, as
illustrated in Figure 6.2. Those problems are commonly solved by performing multiple
laser scans and register the resulting point clouds to stitch them into one (Makadia et al.,
2006). This stitching operation can be time intensive and create errors (Rabbani et al.,
2007). In other scenarios the stitching is favored by marker points that are placed in the
scene (Bornaz & Rinaudo, 2004). In case of excavation sites this is troublesome due to
the constant change of the excavation pit. Additionally, laser scanning requires trained
personnel and expensive equipment (Flood, 2001).
Figure 6.1: Laser scanning an excavation pit. The laser scanner is placed in the right of
the image, illustrated by a gray square. The blue area shows the field of view of the laser
scanner. Due to the shape of the pit, the scanner can hardly capture the entire soil surface.
6.1.2 Photogrammetry
Compared to laser scanning, photogrammetry is performed using a standard digital cam-

era, as described in Section 5.4. A set of photographs is transformed into a point cloud by
matching features among the individual images. A human operator surrounds the scene
while taking photographs at regular intervals. In case of an excavation pit, all visible
areas of the pit are photographed in an overlapping tile manner, as illustrated in Figure
6.3, from different positions surrounding the pit. It is important that the same areas are
captured from multiple perspectives as is illustrated in Figures 6.4 and Figures 6.5. As
all the photos taken by the operator should capture even partly occluded areas, the pho-
togrammetric point cloud generated from the photos covers the entire pit unless there are
Figure 6.2: Laser scanning an excavation pit. The laser scanner is placed in the right of
the image, illustrated by a gray square. The blue area shows the field of view of the laser
scanner. Due to the shape of the pit and an excavator occluding parts of the area, the
scanner can hardly capture the entire soil surface.
areas not visible from any point outside of the pit. Additionally photos from inside the pit
can be taken.
Figure 6.3: The operator’s perspective, looking into the excavation pit, while taking pho-
tos for photogrammetric point cloud generation. The ten blue rectangles represent the
photos taken by the operator. The photos need to overlap such that they share common
features.
Figure 6.6 shows a practical example where photos haven been taken surrounding the
pit and additional pictures were taken from inside the pit. Furthermore the pictures il-
lustrated in the lower right were taken from an elevated angle off a tower crane. After
the images have been processed by the structure from motion algorithm VisualSFM (Wu,
2011), bundle adjustment (Wu et al., 2011), and the patch-based multi-view stereo algo-
rithm (Furukawa & Ponce, 2010) the dense point cloud representation shown in Figure
6.7 was generated.
Figure 6.4: Taking photos of an excavation pit for photgrammetric point cloud generation.
The darker area in the center is the excavation pit. Each blue triangle shows the orientation
of a photo that was taken. The operator moves around the pit taking photos of all currently
visible areas of the pit.
Figure 6.5: Cutting plane of excavation pit while taking photos for photogrammetric point
cloud generation. Each blue triangle shows the orientation of a photo that was taken. At
each location the operator takes photos of all currently visible areas of the pit.
6.1.3 Processing of Point Clouds
After a point cloud is generated by the algorithms described in the previous Section and
in Section 5.4 it contains noise and includes unwanted geometry, created by machines,
as well as structures, and buildings surrounding the excavation pit. Furthermore the in-
dividual points of the cloud need to be post-processed in order to get the desired volume
measure.
6.1.3.1 Clustering
As a first step the point cloud is cleaned from unwanted points outside the region of inter-
est. This is performed using conditional euclidean clustering where points are combined
into different clusters, similar to the approach by Trevor et al. (2013). Each point that is
Figure 6.6: Camera positions of photos taken on an actual construction site. The circular
area is the excavation pit and each pyramid represents a photo that was taken.
added to a cluster needs to be within a certain euclidean distance to another point within
the cluster. The resulting clusters differ in size and the main cluster which forms the ex-
cavation pit can be identified as the largest one. The maximum distance of a point to a
cluster is a parameter for this operation, but can be determined automatically based on the
overall size of the point cloud (Shi et al., 2011). All other clusters are removed from the
point cloud. The result of this process is shown in Figure 6.10. A survey of other possible
approaches has been published by Nguyen & Le (2013).
6.1.3.2 Cleaning
As a next step the point cloud is analyzed by generating a vertical histogram of the points
contained in each layer of the point cloud. The thickness of the layers only has a small
effect on the results, but should be sufficiently large to have points in every layer, but suf-
ficiently small to get the desired histogram information. Taking between 10 and 50 slices
over the full vertical range of the point cloud, depending on the depth of the excavation
pit, has proven to be a good measure. Figure 6.9 shows the histogram for the point cloud
in Figure 6.8. The plot shows that the majority of the points lie in the range between the
elevations 5 and 36. The large spike around an elevation of 30 indicates the bottom of the
excavation pit where most points are located. In case of a non horizontal ground plane,
this spike would be less obvious. The other end of the plot, where the number of points
Figure 6.7: Point cloud generated from photographs illustrated in Figure 6.6.
Figure 6.8: Raw unprocessed point cloud. Structures outside the excavation area have
been included into the point cloud.
flats out at an elevation of 5 marks the upper end of the excavation pit. This mark can be
used for the volume calculation in the following Section. The point cloud resulting from
this procedure is illustrated in Figure 6.10.
40000
30000
Number of points
20000
10000
0
0 10 20 30 40 50
Vertical axis
Figure 6.9: Vertical histogram of point cloud. The graph shows the number of point on
the different elevations in the point cloud. The bottom layer within the point cloud is
clearly visible as the spike at around an elevation of 30. The surrounding ground layer
can be estimated to be at the left of the graph where the histogram flats out.
6.1.3.3 Volume Calculation
In order to calculate the volume of the excavated soil, the correct size of the point cloud
has to be known. When the point cloud was generated by a laser scanner, the size is
known. If however the point cloud was generated by photogrammetry, it needs to be
scaled to the correct size. This can either be done manually by taking a known measure,
such as the length of the excavation pit, an included bored pile wall, or automatically by
placing marker points at known locations. The markers can be located within the point
cloud and the scale of the cloud can then be determined by comparing their distance in
the real world with their distance in the point cloud.
Once the scale of the point cloud is known, there are different ways to calculate the
volume of the excavation pit. The most intuitive way is the create the convex hull of the
point cloud. This can be done using the quickhull algorithm by Barber et al. (1996). In
most cases the convex hull is larger than the actual volume, especially with irregularly
shaped excavation pits, such as the u-shaped pit in Figure 6.11, the error is significant and
cannot be neglected. Figure 6.12 shows the convex hull of the point cloud in Figure 6.10.
Figure 6.10: The point cloud from Figure 6.8 after cleaning. Points outside the excavation
area have been removed.
Figure 6.11: The convex hull of an u-shaped excavation pit would have a much higher
volume than the excavation pit. The white area is the actual excavation pit, the orange
area is the convex hull.
A more advanced approach to determine the volume of an object represented by a

point cloud would be to generate a mesh from the point cloud. Calculating the volume of
a closed mesh is trivial and can be done using signed tetrahedron volumes (Hoppe, 1999).
However, the point cloud does not represent an object with a closed surface in case of an
excavation pit since it is open to the top. Meshing algorithms such as the Poisson surface
reconstruction algorithm by Kazhdan et al. (2006) will close the top opening, creating
a bubble on top of it as is shown in Figure 6.13. The volume of such a mesh does not
represent the volume of the excavation pit. The ball pivoting algorithm by Bernardini
et al. (1999), on the other hand, does not close the top opening at all, as shown in Figure
6.14. It does therefore not allow the calculation of the mesh volume. Additionally the
algorithm can leave holes within the mesh.
Figure 6.12: Convex hull of the point cloud in Figure 6.10. The hull was created using
the quickhull algorithm by Barber et al. (1996).
Figure 6.13: Meshed surface of the point cloud in Figure 6.10. The surface was generated
using the Poisson surface reconstruction algorithm by Kazhdan et al. (2006).
Figure 6.14: Meshed surface of the point cloud in Figure 6.10. The surface was generated
using the the ball pivoting algorithm by Bernardini et al. (1999). The red circle shows a
hole that was left open.
Based on the observation that the walls of an excavation pit hardly allow any overhang,
because overhanging soil would cave in, a more specialized algorithm was developed7 .
The elevation of the surrounding soil is known due to the histogram analysis described in
Section 6.1.3.2.
Under the assumption of walls without overhang the point cloud can be reduced to its
ground layer by removing all vertical duplicates of points, except the lowest ones. De-
pending on the density of the point cloud a tolerance can be defined in order to determine
which points cover each other. Once all vertical duplicates have been removed, the point
cloud is reduced to a ground layer as shown in Figure 6.15.
Figure 6.15: The point cloud from 6.10 after all vertical duplicate points have been re-
moved.
Since the ground layer could be irregular or have gaps, it can be homogenized and
thinned out. First of all all gaps are linearly interpolated, creating a refined ground layer
as is shown in Figure 6.16. Then the cloud can be thinned out to a lower density with
points at regular distances as is shown in Figure 6.17. The ground layer can then be
triangulated into a mesh such that it forms a set of triangles. Each of those triangles is
then extruded towards the elevation of the surrounding soil which was determined by the
histogram analysis. Each triangle creates a prism of which the volume can be calculated.
The sum of all prism volumes adds up to the sum of the excavation pit.
The described method was successfully tested on multiple construction sites to deter-
mine the excavated volume at multiple time steps. Detailed results are described in the
context of multiple case studies in Section 8. Situations where the surrounding ground
is very uneven or at a steep slope might cause the results to decrease in quality. Such
situations need further field studies in order to be evaluated.
7
The idea for this algorithm was developed in close cooperation with Benjamin Steer in the context of his
interdisciplinary project "Processing of Point Clouds of Excavation Sites to obtain Progress Information".
Figure 6.16: The point cloud from Figure 6.15 after refinement to fill holes.
Figure 6.17: The point cloud from Figure 6.16 after resampling to reduce the number of
points to be evenly distributed.
6.1.4 Comparison of Laser-scanning and Photogrammetry
During a comparison study of laser scanning and photogrammetry, several key differences
were determined. The laser scanner needs fewer accessible points on the site, and modern
scanners can capture the point clouds in less than a second (Kusnoto & Evans, 2002). The
laser scanner however is an expensive piece of equipment and requires trained personnel.
For instance, a Leica ScanStation P40 costs more than 100.000EUR. Furthermore it can
be sensitive to weather conditions, such as rain. As a single laser scan can rarely capture
the entire excavation pit, the resulting point clouds need to be stitched. This requires
manual work or the placement of marker points on the site. Additionally laser scanning
is, due to the fewer perspectives that are captured, more sensitive to occlusions as shown
in Figure 6.2. Photogrammetry on the other hand only requires a standard digital camera
6.2 Video-Based Tracking of Excavation Processes 145
and a non professional operator. In the performed field studies, a basic explanation of
the scheme described in Section 6.1.2 sufficed for the personnel to take the right photos.
The approach however requires the area around the excavation pit to be accessible. Large
spans around the pit that are inaccessible can cause the structure from motion approach
to fail. Furthermore puddles of water in the excavation pit create errors. If they are small,
they can be filled in using the approach described in Section 6.1.3.3. Another problem
that may occur is snow. If the pit is covered in snow, the feature extraction of the structure
from motion algorithm will not be able to gather sufficient data to triangulate a dense
point cloud. In summary these limitations did not create major problems during the case
studies.
6.2 Video-Based Tracking of Excavation Processes

In the previous Section an approach to measure the progress of excavation processes was
introduced. However, a detected delay cannot be traced back to the underlying problem
that caused it. Ogunmakin et al. (2013) presented a system to monitor machines involved
in excavation processes and quantify their interactions. The work focuses on tracking
excavators and dumpers in video streams and classify their states. This includes count-
ing the number of dumpers that enter and leave the site as well as the number of bucket
loads an excavator fills into a dumper. This allows the estimation of the excavation pro-
ductivity. However, as the precise volume of a bucket load and of a dumper load cannot
be determined, the approach suffers from a cumulative error when measuring the overall
excavation progress.
The weaknesses of both approaches can be compensated if they are used in synergy8
as is illustrated in Figure 6.18. The cumulative error of the video analysis approach can be
corrected by the absolute volume measurements provided by the photogrammetry, while
the underlying problems of delays can be identified using the video analysis approach.
Aerial
images Excavated
Analysis & control
Photogrammetry
Construction Site
Ground volume
images
Video
Video Activity
analysis statistics
Site layout
Figure 6.18: Overview of the system combining the presented photogrammetry approach
with the video analysis system by Ogunmakin et al. (2013).
8
This synergetic approach was developed in close cooperation with Gbolabo Ogunmakin, Patricio A.
Vela from the Georgia Institute of Technology, André Borrmann, and Jochen Teizer.
As previously described, the photogrammetry approach can be used to generate time

series of the excavated volume of an excavation pit. This information can be used to calcu-
late performance factors for the involved resources as described in Section 2.1. However,
changes in the performance cannot be attributed to the individual machines.
While performance can be defined as the amount of time spent working, a more reli-
able measure of performance is the amount of work performed in a certain time. In case
of excavation processes this commonly translates to the volume of soil removed from the
site per day. When the involved machines are monitored it is possible to detect what ratio
of the time the machines are idling. On the other hand a machine can be working all day
but not be productive. For instance soil could be moved around within the excavation pit
while not dumper is present to speed up the loading process once a dumper arrives at the
site. This however does not increase the volume of the excavation pit. Furthermore the
excavator could be working very slow but still be active all day. In order to get an absolute
measure on the productivity of the entire excavation process the amount of soil removed
from the site is the most accurate metric for this performance factor.
The video analysis approach adds detailed information about the machines states over
time to the system. It distinguishes four different states of the dump trucks: absent, static,
moving, and filling. The individual states are illustrated in Figure 6.19. Absent means that
no dump truck is visible in the video stream. Static means that a dump truck is present,
but not interacting with the excavators. Moving means that the dumper is either moving
in or out of the construction site. Filling means that the excavator is currently filling the
dumper with soil. When a decrease in performance is observed, it can be attributed to
the individual machines based on these states. If absent, the dumpers are unavailable and
a probable cause of the delay is a traffic jam on the way between the excavation and the
dump site. Employing additional dumpers can help solving the problem. If dumpers are in
the static state, hence queuing up on the excavation site, the excavators are the bottleneck
in the process. Deploying additional excavators can solve the problem, given there is
sufficient space.
Figure 6.19: Dump truck state estimates for a video segment of 6 minutes duration and
the activity states in a pie chart (Bügler et al., 2016).
The synergetic approach was successfully tested on two large excavation sites as de-
scribed in detail in Section 8. The main problem of the photogrammetry approach were
missing measurements due to inclement weather. Though, due to the absolute measure-
ments obtained by the volume calculation, the missing data points could be interpolated.
Another problem is that the processing of the image data to generate the point clouds
6.3 Summary 147
and calculate the excavated volume caused a delay of about half a day. The same prob-
lem occurred with the video analysis, where both real-time transfer and processing of the
video data was not possible. On the upside, obtaining measurements of the previous day
in the morning of the next day was feasible and sufficient for decision makers to react
to observed problems. Errors in the video processing were mainly caused by temporary
occlusions such as machines moving in front of each other. This can be addressed by plac-
ing multiple cameras at different locations. Generally, even when occlusions caused high
error rates in the video analysis, the acquired measurements still correlated with the ac-
tual activity and provided accurate estimates of the performance. Future research should
address the use of multiple cameras in conjunction and speeding up video processing to
near real-time.
In summary, the video analysis approach adds both temporal information as well as
resource attribution to the data acquisition approach. In a reactive scheduling system
where delays need to be attributed to processes or involved resources and the respective
performance factors it is vital to obtain accurate metrics on the progress of the involved
processes as well as to determine reasons for observed delays. The photogrammetry ap-
proach in combination with the the video analysis approach proved effective for this pur-
pose.
6.3 Summary
This Chapter introduced two novel methods for the monitoring of excavation processes.
Photogrammetry is the creation of point clouds based on photographs taken of an object.
This approach has been used to generate point clouds of excavation pits of which the
volume could be calculated afterwards. While this process showed to be sensitive to
reflective surfaces and snow, field studies showed that it is a viable approach to monitor
the progress of excavation processes.
However it is not possible to detect underlying reasons for observed delays as the
capturing of such point clouds is only feasible with low temporal resolution. In order
to address this issue a second visual approach, video analysis, was introduced. This ap-
proach allows to quantify the interactions between excavators and dump trucks, measure
idle times and count the number of shovel loads and dump trucks entering and leaving
the site. Combining the two sources of information allows to determine the absolute
productivity of the excavation process as well as determining reasons for periods of low
productivity.
This process results in performance factors that can be updated in regular intervals.
Additionally the variations on productivity can be quantified in terms of uncertainty. Both
the performance factors and the probability distributions they are subject to can be used
to update a reactive simulation model and produce updated schedules taking the newly
arisen circumstances into account.
148 7 Data-driven Reactive Scheduling
7 Data-driven Reactive Scheduling
"For every action, there is an equal and opposite reaction." - Newton et al.
(1850)
This thesis presents the concept of reactive scheduling in the domain of construction
projects consisting of excavation and shoring operations. The presented methods in com-
bination form a synergic approach that is in focus of this Chapter.
7.1 Concept
The overall concept of the data-driven reactive scheduling approach presented in this the-
sis is illustrated in Figure 7.1. First, a model of the construction project is created based
on the specific requirements of the project. This process is described in detail in Section
2.6. The modeling can be done on the basis of process models as discussed in Section
2.5. During this step, the right choice of granularity as described in Section 2.4 and the
inclusion of uncertainty as described in Section 2.2 have to be considered. Including un-
certainty into the model, given the required data is available, allows for more accurate
predictions and can minimize the amount of re-planning required. The model also needs
to include the required resources, precedence constraints and other required elements as
described in Section 2.6.2. For this process Building Information Models can be utilized
as described in Section 2.6.3 to extract elements that need to be built as well as precedence
relationships among them.
Once a model has been created it can be used to generate schedules taking all specified
circumstances into account. Since not every schedule will meet the requirements of the
planner, a metric on how well a schedule suits the purpose is required as described in
terms of objective functions in Section 3.2. The most commonly used metric in research
is the makespan of the entire project, however in real projects costs, resource utilization
and robustness (Section 3.2.1) commonly form a multi-objective optimization problem as
is subject to Section 3.2.2.
After an objective has been defined, schedules can be generated as described in Sec-
tion 3.4. As the underlying Resource-Constrained Project Scheduling Problem (RCPSP)
is considered NP-Hard, heuristic algorithms are of high interest. This thesis introduced
two new heursitic methods for optimizing schedules. The iterative limited-depth local
7.1 Concept 149
Figure 7.1: Flowchart of the overall concept of the reactive scheduling approach. An
initial schedule is built on the basis of a model of the construction project. Once construc-
tion has started, real-time data is acquired. If construction is not yet finished, the model
is updated and it is verified whether the float time of an activity was exceeded. If that is
the case an updated schedule is generated. This process is repeated until construction has
finished.
search algorithm in Section 4.3 and the adaption of the water wave optimization meta-
heuristic optimization algorithm by Zheng (2015), presented in Section 4.2. Both algo-
rithms can utilize the novel pseudo-serial schedule generation scheme, which in turn is
described in Section 4.1. This scheme combines the advantages of both common parallel
and serial schedule generation schemes. Both presented novel scheduling algorithms can
be used to generate an initial schedule based on the developed model as well as update
the schedule once new information has been acquired. Updating a schedule is in essence
the same process as generating a new schedule, except that all activities that have already
been started in the past are locked into place during a schedule update. Furthermore no
new activities can be scheduled into the past.
On basis of the initial schedule the construction project is started. During the course
of construction, data is acquired for the individual activities, updating the model reac-
tively. Excavation activities can be monitored using the presented photogrammetry and
video-analysis approaches in Sections 6.1.2 and 6.2. As a result, delays of activities are
observed and can furthermore give rise to the underlying performance factors of the in-
volved machines as presented in Section 2.1. Performance factors are generally defined
as the amount of work performed over time. The performance factors as well as the ac-
tivity durations are commonly subject to uncertainty which can be estimated based on the
collected data as described in Section 2.2. If for instance an excavator does not perform
as expected because the soil conditions are different than initially measured, the perfor-
mance factor of the machine is decreased. Therefore activities involving the excavator
will be updated by extending their duration respectively.
Once real-time data has been acquired the updated performance factors and accord-
ingly estimated activity durations are verified for exceedance of the float times which can
be calculated using the Dori Algorithm (DA) which is introduced in Section 3.4.6. If the
float time of an activity is exceeded, a re-planning process is initiated, generating a new
schedule based on the newly arisen circumstances. If the uncertainty in the activity dura-
tions has been estimated in terms of probability distributions as discussed in Section 2.2
Monte-Carlo simulation as described in Section 2.2.6 can be used to obtain metrics on
the effects of the uncertainty on the resulting schedules. Furthermore those metrics can
be used to generate particularly robust schedules minimizing the amount of re-planning
required in the future.
7.2 Illustrative Example

Figure 7.2 shows an illustrative example of a schedule of a simple construction project
involving an excavation activity and four shoring activities. While activity A does not
require any resources that are shared with other activities, all other activities require the
same resource and only one instance of that resource is available. The depicted schedule
is optimal for the problem as there is no other schedule with a shorter makespan.
Figure 7.2: Example of a simple construction schedule. The excavation activity A requires
no shared resources, while the shoring activities B through E share the same resource, of
which only one instance is available. Parallel execution of any of the activities B, C, D,
and E is hence prohibited.
Applying the Dori Algorithm (DA) described in Section 3.4.6 provides the informa-
tion that all activities in the schedule have a float time of zero and hence are critical.
7.2 Illustrative Example 151
Figure 7.3: The schedule from Figure 7.2 after a delay of activity A has been observed.
The delay propagated through the remaining schedule because the float time of activity A
has been exceeded. Activities D and E are now scheduled to be executed later.
Figure 7.4: The schedule from Figure 7.2 after a delay of activity A has been observed,
and since the float time was exceeded, a new schedule was generated. Now activity E is
executed during the delay of activity A.
Therefore every delay in the schedule will propagate through the entire remaining sched-
ule and increase the overall makespan by the amount of the delay. It is assumed that
activity A is monitored, and prior to its finish, it is expected that it will take longer than
anticipated.
This expected delay can be determined on the basis of the photogrammetry-based
monitoring approach presented in Section 6.1. The duration of such an excavation activ-
ity can be a product of the performance factors of the involved excavators and dumpers.
When the measured progress falls behind the expectations, the video-based monitoring
approach described in Section 6.2 helps to attribute the delay to either the involved exca-
vators or dumpers. If the decision makers react by deploying additional resources, those
can be accounted for in terms of updated performance factors, when calculating the new
duration of the activity. When no or insufficient counter measures are taken, the result is
a delay of the activity.

The red bar after activity A in Figure 7.3 illustrates such a delay. As no re-planning
has yet been performed the entire schedule would be delayed pushing activities D and
E into the future and prolonging the schedule. As this is expected prior to the start of
activity D it is still possible to change the schedule at this point. The presented concept
advises to generate a new schedule in order to react to the exceeded float time.
Figure 7.4 illustrates a reactive schedule that has been updated after the observation
of the delay of activity A. As a result the effects of the delay have been counteracted and
the new schedule is exactly as long as the initial one from Figure 7.2. Given the longer
duration of activity A, this new schedule is optimal.
However, in many practical cases a constant resource usage is desirable because the
idling of resources creates unnecessary costs. Figure 7.5 shows the resource profile of the
green resource of the initial schedule from Figure 7.2. The green resource is in constant
use throughout the entire schedule.
Figure 7.5: The resource profile for the green resource of the schedule in Figure 7.2.
Figure 7.6 shows the same resource profile for the delayed schedule from Figure 7.3.
The delay prevents activity D from execution and therefore creates a gap in the resource
profile.
7.3 Summary 153
The resource profile in Figure 7.7 corresponds to the reactive schedule from Figure
7.4. It is visible that the resource gap has been closed. In many cases optimizing a sched-
ule with respect to short makespan will also yield a lean resource profile. Additionally
the costs for hiring the resource could be included as an optimization objective, creating
more costs for the schedule containing the gap. However at a certain duration of the gap it
may become worthwhile to make the resource available to other construction sites in the
meantime.
When taking robustness into account, the schedule shown in Figure 7.4 would have
been a better initial schedule as activity A has a float time equal to the duration of activity
E and the observed delay would not have had any effect on the schedule. This illustrates
that scheduling as an multi-objective optimization problem is more practical. Taking
robustness in terms of maximizing float times into account, so for instance using the sum
of all float times as a second objective, would have selected the final schedule as the initial
schedule, as both have equal makespans.
Another possible approach would be to take uncertainties in the activity durations or
the underlying performance factors into account. If it would have been known that activity
A is likely to be delayed, an approach such as Monte-Carlo simulation as described in
Section 2.2.6 would have also shown the schedule in Figure 7.4 as the better initial option.
However, when it is more likely that the sum of the delays of activities B and C is larger
than the delay of A, the schedule in Figure 7.2 would have been an equally good choice.
Since it is very hard to manually make such considerations in case of more complex
schedules it is advantageous to include uncertainty into the project model beforehand,
given the required data is available. When only coarse data is available, methods like
VIBES (AbouRizk et al., 1991), as presented in Section 2.2.1, have been shown to be
effective in estimating the required probability distributions. If detailed sample data is
available from previous projects or other activities, goodness-of-fit tests, as described in
Section 2.2.5 are useful for finding the most suitable probability distribution. Once a
distribution is selected, its parameters can be updated based on newly acquired data in
order to aid the reactive scheduling approach.
7.3 Summary
This Chapter showed how the methods presented throughout this thesis interlock into a
synergic approach. It has been described how the overall process of creating a data-driven
reactive scheduling model for construction projects works and an illustrative example has
been presented. The example illustrated how a delay can cause an optimal schedule to
become suboptimal.
Furthermore it showed how anticipation of delays can prevent the necessity of re-
planning. This can be realized by taking robustness into account in terms of an multi-
objective optimization approach during the generation of the initial schedule. Further-
more, if the uncertainty of activity durations or performance factors is known, performing
Monte-Carlo simulation in order to evaluate the schedule during generation is a viable
approach towards obtaining more robust schedules.

The following Chapter introduces three case studies which have been performed to
show how the concepts described in this thesis can be put into practice.
155
8 Case Studies
"Experience is the only source of knowledge." - Ranganathananda (1990)
This Chapter presents experimental results obtained through case studies that have
been performed on real construction sites. In order to illustrate the feasibility of the
approaches presented in Sections 6.1.2 and 6.2, three case studies have been performed
on different large real-life construction sites. This Chapter describes each of the cases and
discusses the data that was collected and evaluated.
8.1 Office Building

In 2013 a large office building was built in the Erika-Mann-Straße in Munich. For this
site a large pit had to be excavated. The progress of the excavation was monitored for two
weeks in order to evaluate the capabilities of the photogrammetry-based volume calcula-
tion approach that was presented in Section 6.1.2. Figure 8.1 shows a point cloud that was
generated from the site. Figure 8.3 shows the volume that was calculated from the point
clouds. The low change between days 2 and days 8 can also be verified visually from the
point clouds in Figure 8.2. Furthermore days 6 and 7 were a weekend. The Figure shows
all nine point clouds that were generated during the course of the observation.
Figure 8.1: Point cloud of the office building excavation site.
Figure 8.4 shows performance factors that have been calculated based on the volume
measurements. The blue bars attribute the change of each day to the day of measuring
the difference. The red bars are interpolated, assuming constant change has happened
since the last measurement. Figure 8.5 shows the probability density function of a beta
156 8 Case Studies
Day 0 Day 1 Day 2 Day 3 Day 5 Day 8 Day 9 Day 10 Day 12
Figure 8.2: Time series of point clouds of the office building excavation site.
12000
10000
Excavated Volume in m3
8000
6000
4000
2000
0
0 2 4 6 8 10 12
Time in days
Figure 8.3: Volumes of the series of point clouds in Figure 8.2.
distribution that was created based on the sample data. This distribution could be used to
make forecasts on the future performance of the excavation process. Both goodness-of-
fit tests presented in Section 2.2.5 confirm that this probability distribution fits the data.
The mean excavation performance was 1227.5m3 /day. Moving the excavated soil out of
8.1 Office Building 157
the pit was particularly easy due to the ramp, that is visible in Figure 8.1 on the bottom,
leaving the pit towards the right.
5000
Measurements
Interpolation
4000
3000
2000
1000
0
1 2 3 4 5 6 7 8 9 10 11 12
Time in days
Figure 8.4: Performance factors calculated based on the volume calculations in Figure
8.3.
This case study illustrated the capabilities of the photogrammetry based progress mon-
itoring on a rectangular construction site with a large surface area. Furthermore the cal-
culation of performance factors was demonstrated and the uncertainty can successfully be
described using a beta distribution. The resulting data can be used to maintain synchro-
nization of a reactive simulation model that updates construction schedules based on the
newly obtained information.
158 8 Case Studies
1.5
Probability
0.5
0
0 1000 2000 3000 4000
3
Excavated volume in m /day
Figure 8.5: Beta distribution estimated based on the performance factors in Figure 8.4
with α = 1.4427 and β = 2.7839. The mean performance was 1227.5m3 /day.
8.2 Underground Parking Garage 159
8.2 Underground Parking Garage

Due to the lack of parking spaces in downtown Munich, a new underground parking
garage was built. For this purpose a deep round pit had to be excavated. During the
observation period of 36 days, 12000m3 of soil were excavated. Figure 8.6 shows a point
cloud that was created during the excavation process. It is visible that the soil was lifted
out of the pit by an excavator through a step at about half the depth of the pit. This
slowed down the excavation performance compared to the previous site to 429m3 /day.
An additional factor was the traffic situation in Munich made it particularly difficult for
the dumpers to get to the dump site and back. This fact could be noticed by the extended
absence of dumb trucks on the site. The excavation progress over time is plotted in Figure
8.7. The derived performance factors are illustrated in Figure 8.8. The blue bars show
the observed performance attributed to the day of recording the point clouds, while the
red bars spread the volume delta over the workdays since the last recording. The graph in
Figure 8.9 shows the probability density function of the beta distribution estimated from
the performance data. This distribution can be used to make predictions on the excavation
performance since both goodness-of-fit tests presented in Section 2.2.5 confirm that this
probability distribution fits the data.
Figure 8.6: Point cloud of the underground parking garage excavation site. Lifting soil
out of the pit slowed down the excavation progress.
Additionally a video camera was placed during day 26 of the observation period. 4
hours of video were recorded. A point cloud has been created prior to the recording and
a second one after the recording. The event processor, presented in Ogunmakin et al.
160 8 Case Studies
12000
10000
8000
6000
4000
2000
0
0 10 20 30
Time in days
Figure 8.7: Volume plot of the underground parking garage excavation site.
2500
Measurements
Interpolation
2000
1500
1000
500
0
0 10 20 30 36
Time in days
8.7.
(2013), combined the tracking and activity estimation results to generate the statistics
illustrated in Figure 8.10. The statistics were generated by counting excavator bucket
loads and dump trucks entering and leaving the site. In order to measure the performance
8.2 Underground Parking Garage 161
2.5
2
Probability
1.5
0.5
0
0 200 400 600 800 1000 1200 1400
3
of the algorithm the ground truth was manually counted in the video. A total of ntrucks =
22 were detected in the processed video, which exactly matched the ground truth. A total
of nbuckets = 171 bucket loads were detected compared to the ground truth of nbuckets =
177 bucket loads, implying an error of 3.4%. The total volume of the excavator’s bucket
was 2.5m3 . Using the ground truth nbuckets,gt = 177, results in a performance factor of
3 3
τv,gt = 2.5m4h·177 = 110 mh . Using the estimated nbuckets = 171, results in a performance
3
factor of τv = 2.5m4h·171 = 107/f racm3 h. Through the total excavated volume, calculated
on the basis of the point clouds, the corrected volume per bucket can be calculated. Using
3
the ground truth bucket count, this results in vbucketc or,gt = 418m
177
= 2.36m3 , while using
3
the measured data it results in vbucketc or,gt = 418m
171
= 2.44m3 , also yielding an error of
3.4%. The corrected bucket volume accounts for the swell factor of the soil as well as for
the fact that the buckets aren’t always completely filled. From the pie chart in Figure 8.10a
it is visible that the majority of the time, the excavators were waiting for the dumpers9 .
The presented results show that the photogrammetry approach can be used to monitor
the excavation process for long periods of time. Furthermore it shows that the additional
information obtained through the video analysis approach presented by Ogunmakin et al.
(2013) can be used to obtain deeper insight into the progress and to detect underlying
problems causing delays. The precise metrics obtained through this approach can be
used as feedback for a reactive scheduling model either updating performance factors or
estimating updated activity durations and their respective probability distributions.
9
This synergetic approach was developed in close cooperation with Gbolabo Ogunmakin, Patricio A.
Vela, André Borrmann, and Jochen Teizer.
162 8 Case Studies
15%
4% 8 Estimated
Ground Truth
6
Time / min
19% 4
61%
Absent
2
Static
Moving
0
Filling 0 4 8 12 16 22
Truck
(a) Pie chart with aggregate dump truck states (b) Loading time per truck
Figure 8.10: Aggregate dump truck statistics for underground parking garage. The left
part shows a pie chart with the ratios of time spent in the different states. The right part
shows the loading time for the individual trucks (Bügler et al., 2016).
8.3 Hospital Building 163
8.3 Hospital Building

A third construction site was subject to the research presented in this paper. In 2014 the
construction of a large new hospital building with multiple sub-basements was started.
For the purpose of monitoring the excavation progress of the project, six cameras have
been permanently placed around the construction site. With the use of these cameras and
some additional photos taken by on-site personnel point clouds were generated at regular
intervals for a period of 89 days in which more than 20, 000m3 of soil were excavated.
One of the point clouds is visualized in Figure 8.11. The calculated excavation progress
is plotted in Figure 8.12. It is visible, that there were two main excavation phases during
which most of the progres was made. During the flat saddle point around day 20 the bored
and sheet pile walls securing the excavation area were produced and anchored.
Figure 8.11: Point cloud of the hospital excavation site.
Performance factors for the entire period are visualized in Figure 8.13 and the resulting
probability distribution is plotted in Figure 8.14, which was confirmed to fit the data
using the goodness-of-fit test presented in Section 2.2.5. Due to the long recording period
in this case study, the data contains many days on which no excavation processes were
active. Therefore the calculated performance factors do not reflect the actual performance
of the excavation processes in this case. Detailed data on the times when excavation was
idling was not recorded, and therefore exact performance factors could not be calculated a
posteriori. Based on the areas with high activity it can be estimated that the first excavation
3
period between days 4 and 8 had a performance of 1498 md , while the second excavation
3
period between days 35 and 40 had a performance of 808 md . For better calculation of the
performance factors, machine data of the excavators could be used in order to measure
the amount of time they were running. Alternatively a video monitoring approach such
as the one presented in Section 6.2 can be used. Nevertheless this case study showed
that photogrammetry can be used to monitor the progress of large excavation projects
over long periods of time. Given additional machine data or video recordings, precise
calculation of performance factors would be possible.
164 8 Case Studies
20000
15000
10000
5000
0
0 20 40 60 80
Time in days
Figure 8.12: Excavation progress of the hospital excavation site
6000 Measurements
Interpolation
5000
4000
3000
2000
1000
0
0 20 40 60 80 90
Time in days
8.12.
8.4 Summary 165
3.5
2.5
Probability
1.5
0.5
0
0 500 1000 1500 2000
3
8.4 Summary
The results illustrate that the photogrammetry approach to measure the excavated volume
of an excavation pit can be used to monitor an excavation site for long periods of time.
Furthermore performance measures can be calculated and used to make predictions on
the future progress.
The video monitoring approach presented in Section 6.2 was successfully tested and
allowed the exact attribution of delays to underlying circumstances. This information
allows the decision makers to react appropriately to unexpected events. In combination
with the calculated performance factors and their respective probability distributions it is
possible to make accurate predictions for the future. Additionally, the data can be used to
estimate the performance of similar processes on other construction sites.
While the case studies were mostly performed without major problems, some issues
arose during their execution. Bad weather conditions, such as heavy rain, causing large
pools of water, forming large reflective surfaces caused problems when capturing point
clouds using photogrammetry. Since this did not occur during long periods of time and
only single sessions didn’t return a usable point cloud, the problems were practically
neglectable.
In case of the hospital construction site there were long periods of time where no
excavation took place. Those periods need to be marked manually not to be taken into
account when performance factors are updated. Commonly this is happening during the
anchoring of bored and sheet pile walls.
166 9 Summary and Outlook
9 Summary and Outlook
While unexpected delays in large construction projects commonly render plans invalid
and require replanning, taking new circumstances into account, scheduling in construction
industry is still performed in a manual and error-prone way. According to Koehn et al.
(1978) unexpected delays comprise up to 30 percent of the overall project costs.
This thesis presented a concept for the reactive scheduling of large shoring and ex-
cavation projects including new scheduling methods and on-site data acquisition. Ad-
ditionally methods for modeling, simulation, and optimization of construction projects
are introduced. Furthermore a representation of uncertainty is included and methods for
including those methods into the concept are presented.
The main contributions of this thesis are the new pseudo-serial schedule generation
scheme (Section 4.1), the iterative limited-depth tree search algorithm (Section 4.3) for
resource constrained project scheduling problems (RCPSPs), the application of the water-
wave optimization algorithm by Zheng (2015) to RCPSPs (Section 4.2), the uncertainty
extension of the project scheduling library PSPLIB (Section 4.5), as well as the automatic
progress tracking of excavation processes using point clouds (Section 6.1) and its combi-
nation with video-based monitoring (Section 6.2). A number of theoretical and practical
case studies have been performed using the project scheduling library PSPLIB (Section
4.4) and its extenstion, the UPSPLIB (Section 4.5), as well as three different real-life
construction projects which have been accompanied in the course of this thesis (Chapter
8).
Chapter 2 introduced the state-of-the-art in modeling and simulation of construction
operations. A main measure to estimate how long a machine or worker takes to execute a
certain activity are performance factors quantifying the amount of work done in a certain
time (Section 2.1). Those performance factors, and therefore the durations of activities,
are subject to uncertainties which can be modeled using different probability distributions
(Section 2.2). A measure on how good a distribution fits a given set of data can be ac-
quired using goodness-of-fit tests (Section 2.2.5). Furthermore, Monte-Carlo-Simulation
can be used to estimate the consequences of uncertainties in a larger context, such as a
whole construction project (Section 2.2.6). Another, if not the most important aspect of
a construction project is the cost factor (Section 2.3). Costs can be generated in numer-
ous ways and can be static or dynamic. They are created through the use of materials,
resources, workers, or even fines that have to be paid due to missed milestones. A very
important decision to be made when modeling construction operations is the granularity
of the model (Section 2.4).
167
While it is possible to model the precise movement of each machine, such a model
is hard to evaluate and even harder to maintain, when changes occur. Therefore it is
wise to make abstractions where they do not have drastic effect on the outcome. For
instance a two step method can be used. First modeling construction processes, such
as the production of a bored pile wall, as a process model, based on the performance
factors of the involved machines (Section 2.5). The resulting process model can then be
used to obtain an expected duration for the production of that bored pile wall, which then
can be included as one activity into a model of the entire project. For the purpose of
modeling such processes, different models can be found in literature, such as CYCLONE
(Section 2.5.4) or STROBOSCOPE (Section 2.5.5). The resulting construction project
model needs to contain specific information in order to adequately represent the system
to be studied (Section 2.6.2). This information can in part be automatically obtained from
digital representations of the building to be constructed (Section 2.6.3). Once a model has
been successfully created it can be used to perform extensive simulation studies (Section
2.6.4) as well as generate and optimize construction schedules.
Chapter 3 presents a detailed survey on the state-of-the-art in construction schedule
optimization. In any optimization endeavor an objective is of utmost importance (Section
3.2). The possible objectives include costs, makespan, as well as other schedule metrics.
As previously stated, uncertainty is an important aspect of any construction schedule.
Therefore the robustness of a schedule is an important schedule metric (Section 3.2.1).
As often multiple objectives are in the scope of the decision makers, multi-criteria op-
timization methods are of high interest (Section 3.2.2). The project scheduling problem
(PSP) is the problem to schedule a given set of activities under precedence constraints
(Section 3.3). This can be solved using the critical path method (Section 3.3.3) or the pro-
gramm evaluation and review technique (Section 3.3.4). This, however, neglects the lim-
ited availability of resources, which adds considerably more complexity the problem. The
resulting problem is called the resource constrained project scheduling problem (Section
3.4). In order to generate feasible schedules for a given RCPSP, two elementary schedule
generation schemes, the serial and the parallel scheme, can be found in literature (Section
3.4.4). Problems can be evaluated prior to scheduling using different problem character-
istics (Section 3.4.2).
Furthermore schedules can be generated by simulation (Section 3.4.5) such as the
constraint-based discrete-event simulation. Using simulation-based methods allows to
obtain additional measures of the schedules, such as the critical path and float times (Sec-
tion 3.4.6). Generally RCPSPs can be solved in a deterministic way (Section 3.4.7), but
these methods are hardly feasible on a large scale. Therefore heuristic approaches are
subject to extensive research (Section 3.4.8) and numerous methods have been published
in the last decades (Section 3.4.8.9). Additionally Kolisch & Sprecher (1996) published a
project scheduling problem library (PSPLIB) which contains more than 2000 benchmark
problems that are widely used in literature (Section 3.4.9).
Chapter 4 presents the main contributions in the field of construction scheduling of
this thesis. The pseudo-serial schedule generation, as a compromise of the serial and
parallel schedule generation schemes, combines the advantages of both other schemes. It
can generate a superset of the set of active schedules, the set of pseudo-active schedules,
while at the same time being computationally more efficient and having a more continuous
parameter space, that has an increased likelihood of finding good solutions in a confined
area (Section 4.1).
Built upon the new scheduling scheme, the water wave optimization algorithm by
Zheng (2015) was applied to RCPSPs and has been shown to produce results outperform-
ing many other published heuristics (Section 4.2). The performance of the algorithm was
also compared with the results of using the traditional scheduling schemes. The advantage
of the pseudo-serial scheduling scheme was shown to be the adjustability of the search
space, as well as faster convergence of the solution.
Additionally another heuristic algorithm, the iterative swap-based limited depth tree
search was presented (Section 4.3). Compared to the water wave optimization algorithm
the results of this algorithm were worse, but since it starts from a given schedule by
applying swaps and delays, it can be particularly useful for updating available schedules
in the context of reactive re-planning.
The commonly used PSPLIB lacks the availability of problems containing uncertain
activity durations. Therefore it was extended by adding beta distributed delays to each of
the activities of each of the PSPLIB single mode problems. Additionally eight objective
functions have been designed to allow optimization for different aspects of the uncer-
tainties using Monte-Carlo-Simulation (Section 4.5). If newly acquired data changes the
circumstances, durations, performance factors, or probability distributions, an update of
the existing schedule might be required, forming the reactive component of the proposed
approach (Section 4.6).
Experiments performed using the PSPLIB showed clearly that the presented algo-
rithms compare well to other heuristic algorithms in the field of RCPSPs and even out-
perform many of them (Section 4.4) . Furthermore, it has been shown that the adaption
of the water wave optimization to RCPSPs using the pseudo-serial schedule generation
scheme is capable of creating good and robust solutions for the uncertainty extensions of
the PSPLIB (Section 4.5).
Chapter 5 provides an overview of the state-of-the-art in data acquisition on construc-
tion sites. Most commonly, manual protocol-based tracking, such as noting milestones
in spreadsheets, is used for progress monitoring (Section 5.1). Additionally some mod-
ern machines record data on the machine state during operation, which can be used for
progress monitoring (Section 5.2). Another viable data source is laser scanning (Section
5.3). However, due to its need for trained personnel and expensive equipment, modern
advances in the field of photogrammetry have gained increased interest by the commu-
nity (Section 5.4). Additionally radio-based tracking, including GPS, and active, as well
as passive RFID methods, are drawing attention, due to the steadily decreasing costs for
the required hardware (Section 5.6). Only recently, methods of range imaging and video
analysis have become computationally and technically feasible enough to have entered
the realm of construction monitoring research (Section 5.7).
Chapter 6 described advances in the progress monitoring of excavation processes that
were made in the course of this thesis and form additional contributions to the field of con-
struction progress monitoring. In order to track the total volume of excavated soil a point
cloud based method is presented (Section 6.1). The point clouds are obtained through pho-
togrammetry and reduced to the area of interest by smart clustering and cleaning methods.
169
Finally the volume of the resulting representation of an excavation pit is calculated and
can be used as a metric for the excavation progress. As this method only yields the abso-
lute excavated volume, it does not allow to track down the reasons for observed delays.
For this reason the method has been extended by a video-based tracking algorithm which
was developed in synergy with Gbolabo Ogunmakin, Patricio A. Vela from Georgia Tech,
André Borrmann, and Jochen Teizer (Section 6.2). This approach allows to count the
number of trucks entering and leaving the site, as well as the number of buckets loaded
onto the trucks. Furthermore the state of excavators and trucks can be determined. This
allows for detailed analysis of the reason for delays, such as a lack of trucks or excavators.
Chapter 7 summarizes the overall concept presented in this thesis by illustrating how
the individual contributions interlock into a synergic approach for data-driven reactive
scheduling. The concept is furthermore illustrated by an executed example of a simple
schedule. The example shows how re-planning effects the performance of a construction
project and how the modeling of uncertainty and robustness can be utilized in a multi-
objective optimization approach in order to reduce the probability of requiring re-planning
during project execution.
Chapter 8 presents three case studies that have been performed on real construction
sites in order to evaluate the performance of the contributions of this thesis. The case stud-
ies showed that the photogrammetry-based progress monitoring of excavation sites pro-
vides a powerful tool for monitoring excavation processes under different circumstances.
Time series of point clouds were recorded and the progress of excavation could be tracked
precisely. Additionally it was shown, that through the addition of the video-based ap-
proach by Bügler et al. (2016) the underlying reasons for delays could be determined
efficiently. Through the calculation of excavated volume and the detection of machine
states, periods of low performance could be attributed to the involved resources and their
performance factors could be estimated accordingly.
While the presented concept is applicable in the discussed scenarios, it is subject
to several limitations. The presented scheduling algorithms, given scenarios of mod-
ern large scale construction projects involving thousands of activities are still requiring
large amounts of computational power and long periods of time. Finding the right level
of abstraction is vital. However combining small activities to reduce the complexity of
the model may create inaccurate resource allocations. On the other hand the presented
methods will, even in complex cases, be able to return good solutions in acceptable time.
Construction sites that are very constricted and offer little possibility to be photographed
from outside the pit are hard to be monitored using the photogrammetry approach. Fur-
thermore the video analysis concept requires cameras to be on site. Those cameras need
power and a way to transfer the recorded image data to a server for processing. If no wire-
less network or fast mobile network is available, storage cards would need to be retrieved
from the cameras, making the approach rather impractical. The use of unmanned aerial
vehicles (UAV) may be a solution to this limitation.
In practice many construction projects involve different contractors to perform dif-
ferent activities. Once the initial plans are in motion it may not be possible to shift the
activity performed by another contractor in time as the contractor works on his own sched-
ule on different projects. Furthermore different parts such as precast concrete parts might
be produced just in time and be delivered to the site. If there is no space to intermedi-
ately store the delivered parts, shifting that activity may not be possible. In the case of
excavation and shoring operations, the projects are however usually performed by a single
contractor and hardly involve any large just in time delivered parts.
The presented work could be extended by a more detailed representation of the costs of
construction projects. The transportation of additional resources to and from the construc-
tion sites create costs that are currently not reflected in the described representation. The
presented scheduling algorithms could be extended towards to resource leveling problem
(RLP) in order to compare the results with respect to robustness. Additionally the per-
formance factors for excavation processes could be extended. Špačková & Straub (2013)
presented a dynamic Bayesian network approach to model the uncertainty in tunnel exca-
vation based on geological data, taking extraordinary events into account. Such a model
could also be implemented for regular excavation processes. In the process model of ex-
cavation processes, the excavation pit could be split into multiple excavation areas, as is
done in the example in Section 3.1. The capability of calculating the volumes of each
of the excavation areas could be added to the point cloud-based excavation monitoring
approach.
171
10 References
AbouRizk, S. & Mather, K. (2000), Simplifying Simulation Modeling through Integration

with 3D CAD, Journal of Construction Engineering and Management ASCE, 126(6),
475–483.
AbouRizk, S. M. & Halpin, D. W. (1992), Statistical Properties of Construction Duration
Data, Journal of Construction Engineering and Management, 118(3), 525–544, ISSN
0733-9364, doi:10.1061/(ASCE)0733-9364(1992)118:3(525).
AbouRizk, S. M.; Halpin, D. W. & Wilson, J. R. (1991), Visual Interactive Fitting of Beta
Distributions, Journal of Construction Engineering and Management, 117(4), 589–
605, ISSN 0733-9364, doi:10.1061/(ASCE)0733-9364(1991)117:4(589).
AbouRizk, S. M.; Halpin, D. W. & Wilson, J. R. (1994), Fitting Beta Distributions Based
on Sample Data, Journal of Construction Engineering and Management, 120(2), 288–
305, ISSN 0733-9364, doi:10.1061/(ASCE)0733-9364(1994)120:2(288).
Adeli, H. & Karim, A. (1997), Scheduling/Cost Optimization and Neural Dynam-
ics Model for Construction, Journal of Construction Engineering and Management,
123(4), 450–458, ISSN 0733-9364, doi:10.1061/(ASCE)0733-9364(1997)123:4(450).
Agarwal, A.; Colak, S. & Erenguc, S. (2011), A Neurogenetic approach for the resource-
constrained project scheduling problem, Computers & Operations Research, 38(1), 44–
50, ISSN 03050548, doi:10.1016/j.cor.2010.01.007.
Ahmadabadian, A. H.; Robson, S.; Boehm, J.; Shortis, M.; Wenzel, K. & Fritsch, D.
(2013), A comparison of dense matching algorithms for scaled surface reconstruction
using stereo camera rigs, ISPRS Journal of Photogrammetry and Remote Sensing, 78,
157–167, ISSN 09242716, doi:10.1016/j.isprsjprs.2013.01.015.
Alcaraz, J. & Maroto, C. (2001), A Robust Genetic Algorithm for Resource Allocation
in Project Scheduling, Annals of Operations Research, 102, 83–109, doi:10.1023/A:
1010949931021.
Alcaraz, J. & Maroto, C. (2009), Genetic Algorithms for the Resource-Constrained
Project Scheduling Problem, BEIO, Boletín de Estadística e Investigación Operativa,
25(1), 22–31.
172 10 References
Alvarez-Valdes, R.; Crespo, E.; Tamarit, J. & Villa, F. (2008), GRASP and path relinking
for project scheduling under partially renewable resources, European Journal of Opera-
tional Research, 189(3), 1153–1170, ISSN 03772217, doi:10.1016/j.ejor.2006.06.073.
Arbuthnot, J. M. (1712), The History of John Bull.

Artigues, C.; Leus, R. & Talla Nobibon, F. (2013), Robust optimization for resource-
constrained project scheduling with uncertain activity durations, Flexible Ser-
vices and Manufacturing Journal, 25(1-2), 175–205, ISSN 19366582, doi:10.1007/
s10696-012-9147-2.
Assaf, S. A. & Al-Hejji, S. (2006), Causes of delay in large construction projects, In-
ternational Journal of Project Management, 24(4), 349–357, ISSN 02637863, doi:
10.1016/j.ijproman.2005.11.010.
Astrom, K.; Heyden, A.; Kahl, F. & Oskarsson, M. (1999), Structure and Motion from
Lines Under Affine Projections, in Proceedings of the 7th International Conference
on Computer Vision, Kerkyra, Greece, ISBN 0769501648, pp. 285–292, doi:10.1109/
ICCV.1999.791232.
Atkinson, R. (1999), Project Management: Cost, Time and Wuality, Two Best Guesses
and a Phenomenon, its Time to Accept Other Success Criteria, International Journal of
Project Management, 17(6), 337–342, ISSN 02637863, doi:10.1016/S0263-7863(98)
00069-6.
Aziz, R. F.; Hafez, S. M. & Abuel-Magd, Y. R. (2014), Smart optimization for mega con-
struction projects using artificial intelligence, Alexandria Engineering Journal, 53(3),
591–606, ISSN 11100168, doi:10.1016/j.aej.2014.05.003.
Baar, T.; Brucker, P. & Knust, S. (1999), Tabu Search Algorithms and Lower Bounds for
the Resource-Constrained Project Scheduling Problem, in Meta-Heuristics, Springer,
pp. 1–18, doi:10.1007/978-1-4615-5775-3{\_}1.
Ballestín, F.; Barrios, A. & Valls, V. (2011), An evolutionary algorithm for the resource-
constrained project scheduling problem with minimum and maximum time lags, Jour-
nal of Scheduling, 14(4), 391–406, ISSN 10946136, doi:10.1007/s10951-009-0125-9.
Baloi, D. & Price, A. D. (2003), Modelling global risk factors affecting construction
cost performance, International Journal of Project Management, 21(4), 261–269, ISSN
02637863, doi:10.1016/S0263-7863(02)00017-0.
Banks, J.; Carson, J.; Nelson, B. & Nicol, D. (2001), Discrete-Event System Simulation,
Prentice Hall, ISBN 0-13-088702-1.
Barber, C. B.; Dobkin, D. P. & Huhdanpaa, H. (1996), The Quickhull algorithm for con-
vex hulls, ACM Transactions On Mathematical Software, 22(4), 469–483.
173
Bauer, M. (2009), Resource-Constrained Project Scheduling - Models and Extensions,

Ph.D. thesis, University of Hamburg.
Bay, H.; Tuytelaars, T. & Van Gool, L. (2006), SURF: Speeded Up Robust Features, in
Proceedings of the 9th European Conference on Computer Vision, pp. 404–417, doi:
10.1007/11744023{\_}32.
Beis, J. & Lowe, D. (1997), Shape indexing using approximate nearest-neighbour search
in high-dimensional spaces, Proceedings of IEEE Computer Society Conference on
Computer Vision and Pattern Recognition, 1000–1006, ISSN 1063-6919, doi:10.1109/
CVPR.1997.609451.
Beißert, U. (2010), Constraint-basierte Simulation zur Terminplanung von Aus-
führungsprozessen Repräsentation baubetrieblichen Wissens mittels Soft Constraints,
Ph.D. thesis, Bauhaus-Universität Weimar.
Beißert, U.; König, M. & Bargstädt, H.-J. (2007), Constraint-Based Simulation of Outfit-
ting Processes in Building Engineering, in Proceedings of CIB-W78 24th International
Conference on Information Technology in Construction, pp. 491–498.
Beißert, U.; König, M. & Bargstädt, H.-J. (2010), Soft Constraint-based simulation of ex-
ecution strategies in building engineering, Journal of Simulation, 4(4), 222–231, ISSN
1747-7778, doi:10.1057/jos.2010.8.
Bell, C. E. & Han, J. (1991), A New Heuristic Solution Method in Resource-Constrained
Project Scheduling, Naval Research Logistics, 38, 315–331, ISSN 0894069X, doi:10.
1002/1520-6750(199106)38:3<315::AID-NAV3220380304>3.0.CO;2-7.
Bell, C. E. & Park, K. (1990), Solving resource-constrained project scheduling problems
by A* search, Naval Research Logistics, 37(1), 61–84, ISSN 0894069X, doi:10.1002/
1520-6750(199002)37:1<61::AID-NAV3220370104>3.0.CO;2-S.
Benevolenskiy, A.; Roos, K.; Katranuschkov, P. & Scherer, R. J. (2012), Construction pro-
cesses configuration using process patterns, Advanced Engineering Informatics, 26(4),
727–736, ISSN 14740346, doi:10.1016/j.aei.2012.04.003.
Bernardini, F.; Mittleman, J.; Rushmeier, H.; Silva, C. & Taubin, G. (1999), The Ball-
Pivoting Algorithm for Surface Reconstruction, IEEE Transactions on Visualization
and Computer Graphics, 5(4), 349–359, ISSN 10772626, doi:10.1109/2945.817351.
Bershtein, L.; Knyazeva, M. & Rozenberg, I. (2015), Fuzzy resource-constrained project
scheduling for GIS software development, in Proceedings of the 16th World Congress
of the International Fuzzy Systems Association (IFSA) and the 9th Conference of the
European Society for Fuzzy Logic and Technology (EUSFLAT), pp. 1542–1548.
Bettemir, Ö. H. & Sonmez, R. (2012), Hybrid Genetic Algorithm with Simulated An-
nealing for Resource-Constrained Project Scheduling, Journal of Management in En-
gineering, 31(5), 1–8, doi:10.1061/(ASCE)ME.1943-5479.0000323.
174 10 References
Björk, B.-C. (1989), Basic structure of a proposed building product model, Computer-
Aided Design, 21(2), 71–78, doi:10.1016/0010-4485(89)90141-3.
Björk, B.-C. (1992), A Unified Approach for Modeling Construction Information, Build-
ing and Environment, 27(2), 173–194.
Blazewicz, J.; Lenstra, J. & Kan, A. (1983), Scheduling subject to resource constraints:
classification and complexity, Discrete Applied Mathematics, 5(1), 11–24.
Blum, C. & Roli, A. (2003), Metaheuristics in combinatorial optimization: overview and
conceptual comparison, ACM Computing Surveys, 35(3), 189–213, ISSN 02545330,
doi:10.1007/s10479-005-3971-7.
Boctor, F. F. (1990), Some efficient multi-heuristic procedures for resource-constrained
project scheduling, European Journal of Operational Research, 49(1), 3–13, ISSN
03772217, doi:10.1016/0377-2217(90)90116-S.
Boctor, F. F. (1993), Heuristics for scheduling projects with resource restrictions and sev-
eral resource-duration modes, International Journal of Production Research, 31(11),
2547–2558, ISSN 0020-7543, doi:10.1080/00207549308956882.
Boctor, F. F. (1996), Resource-constrained project scheduling by simulated annealing,
International Journal of Production Research, 34(8), 2335–2351, ISSN 0020-7543,
doi:10.1080/00207549608905028.
Bogenstätter, U. (2000), Prediction and optimization of life-cycle costs in early design,
Building Research & Information, 28(5-6), 376–386, ISSN 0961-3218, doi:10.1080/
096132100418528.
Bornaz, L. & Rinaudo, F. (2004), Terrestrial laser scanner data processing, Proceedings
of XX ISPRS Commission V Congress, 45(B5), 514–519.
Bosché, F. (2010), Automated recognition of 3D CAD model objects in laser scans and
calculation of as-built dimensions for dimensional compliance control in construction,
Advanced Engineering Informatics, 24(1), 107–118, ISSN 14740346, doi:10.1016/j.
aei.2009.08.006.
Bosché, F.; Ahmed, M.; Turkan, Y.; Haas, C. T. & Haas, R. (2015), The value of integrat-
ing Scan-to-BIM and Scan-vs-BIM techniques for construction monitoring using laser
scanning and BIM: The case of cylindrical MEP components, Automation in Construc-
tion, 49, 201–213, ISSN 09265805, doi:10.1016/j.autcon.2014.05.014.
Bosche, F. & Haas, C. (2008), Automated retrieval of 3D CAD model objects in con-
struction range images, Automation in Construction, 17(4), 499–512, ISSN 09265805,
doi:10.1016/j.autcon.2007.09.001.
Böttcher, J. & Drexl, A. (1999), Project scheduling under partially renewable resource
constraints, Management Science, 45(4), 543–559.
175
Bouleimen, K. & Lecocq, H. (2003), A new efficient simulated annealing algorithm for
the resource-constrained project scheduling problem and its multiple mode version,
European Journal of Operational Research, 149(2), 268–281, ISSN 03772217, doi:
10.1016/S0377-2217(02)00761-0.
Brandt, S. & Bode, M. (2014), Ablaufplanung unter Berücksichtigung von Unschärfe, in
Forum Bauinformatik 2014.
Braun, A.; Tuttas, S.; Borrmann, A. & Stilla, U. (2015a), A concept for automated con-
struction progress monitoring using BIM-based geometric constraints and photogram-
metric point clouds, Journal of Information Technology in Construction, 20(8), 68–79.
Braun, A.; Tuttas, S.; Borrmann, A. & Stilla, U. (2015b), Automated progress moni-
toring based on photogrammetric point clouds and precedence relationship graphs, in
The 32nd International Symposium on Automation and Robotics in Construction and
Mining (ISARC).
Brooks, G. H. & White, C. R. (1965), An algorithm for finding optimal or near opti-
mal solutions to the production scheduling problem, Journal of Industrial Engineering,
16(1), 34.
Brooks, R. J. & Tobias, a. M. (1996), Choosing the best model: Level of detail, com-
plexity, and model performance, Mathematical and Computer Modelling, 24(4), 1–14,
ISSN 08957177, doi:10.1016/0895-7177(96)00103-3.
Brown, D. C. (1976), The bundle adjustment—progress and prospects, Int. Archives Pho-
togrammetry, 21(3), 1.
Browning, T. R. & Yassine, A. a. (2010), Resource-constrained multi-project scheduling:
Priority rule performance revisited, International Journal of Production Economics,
126(2), 212–228, ISSN 09255273, doi:10.1016/j.ijpe.2010.03.009.
Bügler, M. & Borrmann, A. (2014), Using Swap-Based Search Trees to obtain Solu-
tions for Resource Constrained Project Scheduling Problems, in Proceedings in Applied
Mathematics and Mechanics, doi:10.1002/pamm.201410385.
Bügler, M.; Dori, G. & Borrmann, A. (2013), Swap Based Process Schedule Optimization
using Discrete-Event Simulation, in Proceedings of the International Conference on
Construction Applications of Virtual Reality, doi:10.13140/RG.2.1.2008.8162.
Bügler, M.; Ogunmakin, G.; Vela, P. A.; Borrmann, A. & Teizer, J. (2016), Fusion of Pho-
togrammetry and Video Analysis for Productivity Assessment of Earthwork Processes
(Accepted), Computer-Aided Civil and Infrastructure Engineering.
Błażewicz, J.; Domschke, W. & Pesch, E. (1996), The job shop scheduling problem:
Conventional and new solution techniques, European Journal of Operational Research,
93(1), 1–33, ISSN 03772217, doi:10.1016/0377-2217(95)00362-2.
176 10 References
Calhoun, K. M.; Deckro, R. F.; Moore, J. T.; Chrissis, J. W. & Van Hove, J. C. (2002),
Planning and re-planning in project and production scheduling, Omega, 30(3), 155–
170, ISSN 03050483, doi:10.1016/S0305-0483(02)00024-5.
Cameron, W. B. (1963), Informal sociology: A casual introduction to sociological think-
ing, Random House New York.
Chahrour, R. (2006), Integration von CAD und Simulation auf Basis von Produktmod-
ellen im Erdbau.
Chahrour, R. & Franz, V. (2006), Seamless data model for a CAD-based simulation sys-
tem, in Proceedings of the Joint International Conference on Computing and Decision
Making in Civil and Building Engineering, pp. 3958–3967.
Chan, W.; Chua, D. & Kannan, G. (1996), Construction resource scheduling with genetic
algorithms, Journal of Construction Engineering and Management, 122(2), 125–132.
Chassein, A. & Goerigk, M. (2016), A bicriteria approach to robust optimization, Com-
puters & Operations Research, 66, 181–189, ISSN 03050548, doi:10.1016/j.cor.2015.
08.007.
Chen, R. M.; Wu, C. L.; Wang, C. M. & Lo, S. T. (2010), Using novel parti-
cle swarm optimization scheme to solve resource-constrained scheduling problem in
PSPLIB, Expert Systems with Applications, 37(3), 1899–1910, ISSN 09574174, doi:
10.1016/j.eswa.2009.07.024.
Cheng, M.-Y. & Chen, J.-C. (2002), Integrating barcode and GIS for monitoring con-
struction progress, Automation in Construction, 11(1), 23–33, ISSN 09265805, doi:
10.1016/S0926-5805(01)00043-7.
Cheng, T.; Venugopal, M.; Teizer, J. & Vela, P. (2011), Performance evaluation of
ultra wideband technology for construction resource location tracking in harsh en-
vironments, Automation in Construction, 20(8), 1173–1184, ISSN 09265805, doi:
10.1016/j.autcon.2011.05.001.
Cheok, G. S. & Stone, W. C. (1999), Non-intrusive scanning technology for construction
assessment, in Proceedings of the 16th International Symposium on Automation and
Robotics in Construction, pp. 645–650.
Cheok, G. S.; Stone, W. C.; Lipman, R. R. & Witzgall, C. (2000), Ladars for construction
assessment and update, Automation in Construction, 9(5-6), 463–477, ISSN 09265805,
doi:10.1016/S0926-5805(00)00058-3.
Chtourou, H. & Haouari, M. (2008), A two-stage-priority-rule-based algorithm for robust
resource-constrained project scheduling, Computers & Industrial Engineering, 55(1),
183–194, ISSN 03608352, doi:10.1016/j.cie.2007.11.017.
Clark, W.; Polakov, W. N. & Trabold, F. W. (1922), The Gantt chart: A working tool of
management, The Ronald Press Company.
177
Cooper, D. F. (1976), Heuristics for Scheduling Resource-Constrained Projects: An Ex-

perimental Investigation, Management Science, 22(11), 1186–1194, ISSN 0025-1909,
doi:10.1287/mnsc.22.11.1186.
Costin, A.; Pradhananga, N. & Teizer, J. (2012), Leveraging passive RFID technology
for construction resource field mobility and status monitoring in a high-rise renovation
project, Automation in Construction, 24, 1–15, ISSN 09265805, doi:10.1016/j.autcon.
2012.02.015.
Daum, S. & Borrmann, A. (2014), Processing of Topological BIM Queries using Bound-
ary Representation Based Methods, Advanced Engineering Informatics, 28(4), 272–
286, doi:10.1016/j.aei.2014.06.001.
Davis, E. (1973), Project scheduling under resource constraints - historical review
and categorization of procedures, AIIE Transactions, 5(4), 37–41, doi:10.1080/
05695557308974916.
Davis, E. W. & Patterson, J. H. (1975), A Comparison of Heuristic and Optimum Solu-
tions in Resource-Constrained Project Scheduling, Management Science, 21(8), 944–
955, ISSN 0025-1909, doi:10.1287/mnsc.21.8.944.
Deb, K. & Gupta, H. (2006), Introducing robustness in multi-objective optimization, Evo-
lutionary computation, 14(4), 463–494, ISSN 1063-6560, doi:10.1162/evco.2006.14.4.
463.
Debels, D.; De Reyck, B.; Leus, R. & Vanhoucke, M. (2006), A hybrid scatter
search/electromagnetism meta-heuristic for project scheduling, European Journal of
Operational Research, 169(2), 638–653, ISSN 03772217, doi:10.1016/j.ejor.2004.08.
020.
Debels, D. & Vanhoucke, M. (2005), A Bi-population Based Genetic Algorithm for the
Resource-Constrained Project Scheduling Problem, in Artificial Evolution, Springer
Berlin Heidelberg, pp. 378–387, doi:10.1007/11740698{\_}23.
Demeulemeester, E. L. & Herroelen, W. S. (1997), New Benchmark Results for the
Resource-Constrained Project Scheduling Problem, Management Science, 43(11),
1485–1492, ISSN 0025-1909, doi:10.1287/mnsc.43.11.1485.
Dore, C. & Murphy, M. (2014), Semi-automatic generation of as-built BIM façade ge-
ometry from laser and image data, Journal of Information Technology in Construction,
19(January), 20–46, ISSN 14036835.
Dori, G. (2016), Simulation-based methods for float time determination and schedule
optimization for construction projects, Ph.D. thesis, Technische Universität München.
Dori, G.; Borrmann, A.; Szczesny, K.; Hamm, M. & König, M. (2012), Combining
forward and backward process simulation for generating and analysing construction
schedules, in Proceedings of the 14th Int. Conf. on Computing in Civil and Building
Engineering, Moscow, Russia.
178 10 References
Dorigo, M.; Maniezzo, V. & Colorni, A. (1996), Ant system: Optimization by a colony
of cooperating agents, IEEE Transactions on Systems, Man, and Cybernetics, Part B:
Cybernetics, 26(1), 29–41, ISSN 10834419, doi:10.1109/3477.484436.
Dossick, C. S. & Neff, G. (2010), Organizational Divisions in BIM-Enabled Commercial
Construction, Journal of Construction Engineering and Management, 136, 459–467,
ISSN 0733-9364, doi:10.1061/(ASCE)CO.1943-7862.0000109.
Drexl, a. (1991), Scheduling of Project Networks by Job Assignment, Management Sci-
ence, 37(12), 1590–1602, ISSN 0025-1909, doi:10.1287/mnsc.37.12.1590.
Dzeng, R.-J. & Tommelein, I. D. (2004), Product modeling to support case-based con-
struction planning and scheduling, Automation in Construction, 13(3), 341–360, ISSN
09265805, doi:10.1016/j.autcon.2003.10.002.
Echeverry, D. (1996), Adaptation of barcode technology for construction project control,
in Proceedings of the ASCE Conference on Computing in Civil Engineering, pp. 1034–
1040.
El-Omari, S. & Moselhi, O. (2011), Integrating automated data acquisition technolo-
gies for progress reporting of construction projects, Automation in Construction, 20(6),
699–705, ISSN 09265805, doi:10.1016/j.autcon.2010.12.001.
Elsayed, E. A. (1982), Algorithms for project scheduling with resource constraints, In-
ternational Journal of Production Research, 20(1), 95–103, ISSN 0020-7543, doi:
10.1080/00207548208947751.
Feng, C.-W.; Liu, L. & Burns, S. a. (1997), Using Genetic Algorithms to Solve Construc-
tion Time-Cost Trade-Off Problems, Journal of Computing in Civil Engineering, 11(3),
184–189, ISSN 0887-3801, doi:10.1061/(ASCE)0887-3801(1997)11:3(184).
Fente, J.; Knutson, K. & Schexnayder, C. (1999), Defining a Beta distribution function
for construction simulation, in Proc. of the 1999 Winter Simulation Conference, ISBN
0-7803-5780-9, ISSN 02750708, pp. 1010–1015, doi:10.1109/WSC.1999.816813.
Fischer, M. & Aalami, F. (1996), Scheduling with Computer-Interpretable Construction
Method Models, Construction Engineering and Management, 122(4), 337–347, ISSN
0733-9364, doi:10.1061/(ASCE)0733-9364(1996)122:4(337).
Fishman, G. S. & Kiviat, P. J. (1967), The statistics of discrete-event simulation, Simula-
tion, 10(4), 185–195.
Flachs, G. M.; Thompson, W. E.; Black, R. J.; Taylor, J. M.; Cannon, W.; Rogers, R.
& Others (1977), An automatic video tracking system, in Proceedings of the National
Aerospace and Electronics Conference, pp. 361–368.
Flachs, G. M.; Thompson, W. E.; Gilbert, A. L. & Others (1976), A real-time Struc-
tural Tracking Algorithm, in Proceedings of the National Aerospace and Electronics
Conference, pp. 161–168.
179
Fleszar, K. & Hindi, K. S. (2004), Solving the resource-constrained project scheduling

problem by a variable neighbourhood search, European Journal of Operational Re-
search, 155(2), 402–413, ISSN 03772217, doi:10.1016/S0377-2217(02)00884-6.
Flood, M. (2001), LiDAR activities and research priorities in the commercial sector, In-
ternational Archives of Photogrammetry, Remote Sensing and Spatial Information Sci-
ences, XXXIV(Part 3), 22–24, ISSN 1098-6596, doi:10.1017/CBO9781107415324.
004.
Fu, N.; Lau, H. C.; Varakantham, P. & Xiao, F. (2012), Robust local search for solving
RCPSP/max with durational uncertainty, Journal of Artificial Intelligence Research,
43, 43–86, ISSN 10769757, doi:10.1613/jair.3424.
Furukawa, Y. & Ponce, J. (2010), Accurate, dense, and robust multiview stereopsis, IEEE
Transactions on Pattern Analysis and Machine Intelligence, 32(8), 1362–76, ISSN
1939-3539, doi:10.1109/TPAMI.2009.161.
Gantt, H. L. (1903), A Graphical Daily Balance in Manufacture, ASME Transactions,
24(1), 1322–1336.
Ghoddousi, P.; Eshtehardian, E.; Jooybanpour, S.; Javanmardi, A. & Ghoddousia, P.
(2013), Multi-mode resource-constrained discrete time–cost-resource optimization in
project scheduling using non-dominated sorting genetic algorithm, Automation in Con-
struction, 30, 216–227, ISSN 09265805, doi:10.1016/j.autcon.2012.11.014.
Gilbert, A. L.; Giles, M. K.; Flachs, G. M.; Rogers, R. B. & U, Y. E. E. H. (1980), A Real-
Time Video Tracking System, IEEE Transactions on Pattern Analysis and Machine
Intelligence, 2(1), 47–56, ISSN 0162-8828, doi:10.1109/TPAMI.1980.4766969.
Glover, F. (1989), Tabu Search - Part I, ORSA Journal on Computing, 1(3), 190–206,
ISSN 0899-1499, doi:10.1287/ijoc.1.3.190.
Glover, F. (1990a), Tabu Search - Part II, ORSA Journal on Computing, 2(1), 4–32.
Glover, F. (1990b), Tabu search: A tutorial, Interfaces, 20(4), 74–94.
Golenko-Ginzburg, D. & Gonik, A. (1997), Stochastic network project scheduling with
non-consumable limited resources, International Journal of Production Economics,
48(1), 29–37.
Gonçalves, J.; Mendes, J. & Resende, M. (2008), A genetic algorithm for the resource
constrained multi-project scheduling problem, European Journal of Operational Re-
search, 189(3), 1171–1190, ISSN 03772217, doi:10.1016/j.ejor.2006.06.074.
Goncharov, E. (2011), A greedy heuristic approach for the Resource-Constrained Project
Scheduling Problem, Studia Informatica Universalis, 9(3), 79–90.
180 10 References
Gong, J. & Caldas, C. H. (2010), Computer Vision-Based Video Interpretation Model

for Automated Productivity Analysis of Construction Operations, Journal of Comput-
ing in Civil Engineering, 24(3), 252–264, ISSN 0887-3801, doi:10.1061/(ASCE)CP.
1943-5487.0000027.
Gonsalves, R. & Teizer, J. (2009), Human motion analysis using 3D range imaging tech-
nology, Proceedings of the 26th International Symposium on Automation and Robotics
in Construction, 76–85.
Gordon, G. (1961), A general purpose systems simulation program, in Proc. of the AFIPS
’61 - Eastern Joint Computer Conference: Computers - Key to Total Systems Control,
pp. 87–104.
Grima, M. a.; Bruines, P. a. & Verhoef, P. N. W. (2000), Modeling tunnel boring machine
performance by neuro-fuzzy methods, Tunnelling and Underground Space Technology,
15(3), 259–269, ISSN 08867798, doi:10.1016/S0886-7798(00)00055-9.
Halpin, D. & Martinez, L. (1999), Real world applications of construction process simu-
lation, in Proceedings of the 31st Winter Simulation Conference.
Halpin, D. W. (1977), CYCLONE: Method for modeling of job site processes, Journal of
the Construction Division, 3(103), 489–499.
Halpin, D. W. & Riggs, L. S. (1992), Planning and Analysis of Construction Operations,
John Wiley & Sons, ISBN 047155510X.
Halpin, D. W. & Senior, B. A. (2010), Construction management, John Wiley & Sons,
ISBN 0471787477.
Hartley, R. & Zisserman, A. (2002), Multiple view geometry in computer vision, Cam-
bridge University Press, ISBN 0521540518.
Hartmann, S. (1998), A Competitive Genetic Algorithm for Resource-Constrained Project
Scheduling, Naval Research Logistics (NRL), 45, 733–750, ISSN 1520-6750, doi:10.
1002/(SICI)1520-6750(199810)45:7<733::AID-NAV5>3.0.CO;2-C.
Hartmann, S. & Kolisch, R. (2000), Experimental evaluation of state-of-the-art heuris-
tics for the resource-constrained project scheduling problem, European Journal of Op-
erational Research, 127(2), 394–407, ISSN 03772217, doi:10.1016/S0377-2217(99)
00485-3.
Hashash, Y. M. A.; Oliveira Filho, J. N.; Su, Y. Y. & Liu, L. Y. (2005), 3D laser Scanning
for Tracking supported Excavation Construction, Geo-Frontiers, 24–26, doi:10.1061/
40785(164)2.
Hendrickson, C.; Zozaya-Gorostiza, C.; Rehak, D.; Baracco-Miller, E. & Lim, P. (1987),
An expert system for construction planning, Journal of Computing in Civil Engineer-
ing, 1(4), 253–269, ISSN 08873801, doi:10.1061/(ASCE)0887-3801(1987)1:4(253).
181
Herroelen, W. & Leus, R. (2004), Robust and reactive project scheduling: a review
and classification of procedures, International Journal of Production Research, 42(8),
1599–1620.
Holland, J. H. (1975), Adaptation in Natural and Artificial Systems, The University of
Michigan Press.
Hoos, H. H. & Stützle, T. (2004), Stochastic local search: Foundations & applications,
Elsevier.
Hoppe, H. (1999), New quadric metric for simplifying meshes with appearance attributes,
Proceedings Visualization ’99 (Cat. No.99CB37067), ISSN 1, doi:10.1109/VISUAL.
1999.809869.
Horenburg, T.; Wimmer, J. & Günthner, W. A. (2012), Resource Allocation in Construc-
tion Scheduling based on Multi-Agent Negotiation, in Proceedings of the 14th Int.
Conference on Computing in Civil & Building Engineering.
Huang, R.-y. & Hsieh, B.-c. (2012), System Modeling of Object-Oriented Construction
Process Simulations, in Proceedings of the 29th ISARC, pp. 1–5.
Huhnt, W. & Enge, F. (2006), Can algorithms support the specification of construction
schedules, Journal of Information Technology in Construction, 11, 547–564, ISSN
14006529.
Jackson, J. (1957), Simulation research on job shop production, Naval Research Logistics
Quarterly, 4(4), 287–295.
Jaselskis, E.; Anderson, M. R.; Jahren, C. T.; Rodriguez, Y. & Njos, S. (1995), Radio-
frequency identification applications in construction industry, Journal of Construction
Engineering and Management, 121(2), 189– 196.
Jaśkowski, P. & Sobotka, A. (2006), Scheduling construction projects using evolutionary
algorithm, Journal of Construction Engineering and Management, 132(August), 861–
870.
Jedrzejowicz, P. & Ratajczak-Ropel, E. (2014), Reinforcement Learning strategies for A-
Team solving the Resource-Constrained Project Scheduling Problem, Neurocomputing,
146, 301–307, ISSN 09252312, doi:10.1016/j.neucom.2014.05.070.
Józefowska, J.; Mika, M. & Różycki, R. (2001), Simulated annealing for multi-mode
resource-constrained project scheduling, Annals of Operations Research, 102(1-4),
137–155.
Kagioglou, M.; Cooper, R.; Aouad, G. & Sexton, M. (2000), Rethinking Construction:
The Generic Design and Construction Process Protocol, Engineering Construction and
Architectural Management, 7(2), 141–153, ISSN 0969-9988, doi:10.1108/eb021139.
Kahlmann, T.; Remondino, F. & Guillaume, S. (2007), Range Imaging Technology: New
Developments and applications for people identification and tracking, in Proceedings of
the SPIE - The International Society for optical Engineering, doi:10.1117/12.702512.
182 10 References
Kargul, A.; Bügler, M.; Borrmann, A. & Günthner, W. A. (2015), Web based field data
analysis and data-driven simulation application for construction performance predic-
tion, Journal of Information Technology in Construction, 20, 479–494.
Kazhdan, M.; Bolitho, M. & Hoppe, H. (2006), Poisson surface reconstruction, in Pro-
ceedings of the fourth Eurographics symposium on Geometry processing, Eurographics
Association, Aire-la-Ville, Switzerland, Switzerland, SGP ’06, ISBN 3-905673-36-3,
pp. 61–70.
Kelley, J. & Walker, M. (1961), Critical-Path Planning and Scheduling: Mathematical
Basis, Operations Research, 9(3), 296–320.
Kern, F. (2002), Precise determination of volume with terestical 3D-laserscanner, in
Geodesy for Geotechnical and Structural Engineering II, pp. 531–534.
Kim, K.; Yun, Y.; Yoon, J.; Gen, M. & Yamazaki, G. (2005), Hybrid genetic algorithm
with adaptive abilities for resource-constrained multiple project scheduling, Computers
in Industry, 56(2), 143–160, ISSN 01663615, doi:10.1016/j.compind.2004.06.006.
Kiran, A. S. & Smith, M. L. (1984), Simulation studies in job shop scheduling - A
Survey, Computers & Industrial Engineering, 8(2), 87–93, ISSN 03608352, doi:
10.1016/0360-8352(84)90002-0.
Kirkpatrick, S.; Gelatt, C. D. & Vecchi, M. P. (1983), Optimization by Simulated An-
nealing, Science, 220(4598), 671–680, ISSN 00368075, doi:10.1126/science.220.4598.
671.
Kiziltas, S.; Burcu, A.; Ergen, E. & Pingbo, T. (2008), Technological assessment and
process implications of field data capture technologies for construction and facil-
ity/infrastructure management, ITcon, 13, 134–154.
Klaubert, C. & Günthner, W. A. (2011), Entwicklung eines RFID-basierten Informations-
und Kommunikationssystems für die Baulogistik, Ph.D. thesis.
Klaubert, C.; Schorr, M. & Günthner, W. A. (2010), Real time construction progress
control using NFC, ITG-Fachbericht - RFID Systech.
Klein, R. (2000), Bidirectional planning: Improving priority rule-based heuristics for
scheduling resource-constrained projects, European Journal of Operational Research,
127(3), 619–638, ISSN 03772217, doi:10.1016/S0377-2217(99)00347-1.
Klein, R. W. & Baris, P. M. (1991), Selecting and generating variates for modeling service
times, Computers & Industrial Engineering, 20(1), 27–33.
Kochetov, Y. & Stolyar, A. (2003), Evolutionary local search with variable neighborhood
for the resource constrained project scheduling problem, in Proceedings of the Work-
shop on Computer Science and Information Technologies CSIT’2003.
Koehn, E.; Young, R.; Kuchar, J. & Seling, F. (1978), Cost of delays in construction,
Journal of the Construction Division, 104(3), 323–331.
183
Kolisch, R. (1996a), Efficient priority rules for the resource-constrained project schedul-
ing problem, Journal of Operations Management, 14(3), 179–192, ISSN 02726963,
doi:10.1016/0272-6963(95)00032-1.
Kolisch, R. (1996b), Serial and parallel resource-constrained project scheduling methods
revisited: Theory and computation, European Journal of Operational Research, 90(2),
320–333, doi:10.1016/0377-2217(95)00357-6.
Kolisch, R. & Drexl, A. (1996), Adaptive search for solving hard project scheduling prob-
lems, Naval Research Logistics, 43(1), 23–40, ISSN 0894069X, doi:10.1002/(SICI)
1520-6750(199602)43:1<23::AID-NAV2>3.3.CO;2-4.
Kolisch, R. & Hartmann, S. (1999), Heuristic algorithms for the resource-constrained
project scheduling problem: Classification and computational analysis, in Weglarz, J.
(ed.), Handbook on Recent Advances in Project Scheduling, Springer, ISBN 978-1-
4613-7529-6, pp. 147–178, doi:10.1007/978-1-4615-5533-9{\_}7.
Kolisch, R. & Hartmann, S. (2006), Experimental investigation of heuristics for resource-
constrained project scheduling: An update, European Journal of Operational Research,
174(1), 23–37, ISSN 03772217, doi:10.1016/j.ejor.2005.01.065.
Kolisch, R. & Sprecher, A. (1996), Project Scheduling Problem Library (PSPLIB).
Kolisch, R. & Sprecher, A. (1997), PSPLIB - A project scheduling problem library,
European Journal of Operational Research, 96(1), 205–216, ISSN 03772217, doi:
10.1016/S0377-2217(96)00170-1.
König, M. (2011a), Generation of robust construction schedules using evolution strate-
gies, in Proceedings of the 2011 ASCE International Conference, 2005, pp. 1–8.
König, M. (2011b), Robust construction scheduling using discrete-event simulation, in
Proceedings of the 2011 ASCE International Workshop on Computing in Civil En-
gineering, ASCE, American Society of Civil Engineers, pp. 446–453, doi:10.1061/
41182(416)55.
König, M. & Beißert, U. (2009), Construction scheduling optimization by simulated an-
nealing, in Proceedings of ASCE Workshop on Computing in Civil Engineering, Austin,
TX, pp. 183–190.
König, M.; Beißert, U.; Steinhauer, D. & Bargstädt, H.-J. (2007), Constraint-based Simu-
lation of Outfitting Processes in Shipbuilding and Civil Engineering, in Proceedings of
the 6th EUROSIM Congress on Modeling and Simulation.
König, M.; Koch, C.; Habenicht, I. & Spieckermann, S. (2012), Intelligent BIM-based
construction scheduling using discrete event simulation, in Proceedings of the 2012
Winter Simulation Conference, ISBN 9781467347808, pp. 1219–1229.
184 10 References
Koulinas, G.; Kotsikas, L. & Anagnostopoulos, K. (2014), A particle swarm optimization

based hyper-heuristic algorithm for the classic resource constrained project schedul-
ing problem, Information Sciences, 277, 680–693, ISSN 00200255, doi:10.1016/j.ins.
2014.02.155.
Kroese, D. P.; Brereton, T.; Taimre, T. & Botev, Z. I. (2014), Why the Monte Carlo method
is so important today, Wiley Interdisciplinary Reviews: Computational Statistics, 6(6),
386–392, ISSN 19395108, doi:10.1002/wics.1314.
Kurtulus, I. & Davis, E. W. (1982), Multi-Project Scheduling: Categorization of Heuristic
Rules Performance, Management Science, 28(2), 161–172, ISSN 00251909, doi:10.
1287/mnsc.28.2.161.
Kusnoto, B. & Evans, C. A. (2002), Reliability of a 3D surface laser scanner for or-
thodontic applications, American Journal of Orthodontics and Dentofacial Orthope-
dics, 122(4), 342–348, ISSN 08895406, doi:10.1067/mod.2002.128219.
Law, K. H. & Jouaneh, M. K. (1986), Data modeling for building design, in Proceedings
of the fourth Conference on Computing in Civil Engineering, New York, ASCE, pp.
21–36.
Lenstra, J. & Rinnooy Kan, A. (1979), Computational complexity of discrete optimization
problems, Annals of Discrete Mathematics, 4, 121–140, ISSN 0167-5060.
Leon, V. J. & Balakrishnan, R. (1995), Strength and adaptability of problem-space based
neighborhoods for resource-constrained scheduling, OR Spektrum, 17(2), 173–182,
ISSN 0171-6468, doi:10.1007/BF01719262.
Leung, S.-w.; Mak, S. & Lee, B. L. (2008), Using a real-time integrated communica-
tion system to monitor the progress and quality of construction works, Automation in
Construction, 17(6), 749–757, ISSN 09265805, doi:10.1016/j.autcon.2008.02.003.
Levin, A.; Fergus, R.; Durand, F. & Freeman, W. T. (2007), Image and depth from a
conventional camera with a coded aperture, ACM Transactions on Graphics, 26(3), 70,
ISSN 07300301, doi:10.1145/1276377.1276464.
Li, H.; Chen, Z.; Yong, L. & Kong, S. C. (2005), Application of integrated GPS and
GIS technology for reducing construction waste and improving construction efficiency,
Automation in Construction, 14(3), 323–331, ISSN 09265805, doi:10.1016/j.autcon.
2004.08.007.
Li, H. & Love, P. (1997), Using Improved Genetic Algorithms to Facilitate Time-Cost
Optimization, Journal of Construction Engineering and Management, 123(3), 233–
237, ISSN 0733-9364, doi:10.1061/(ASCE)0733-9364(1997)123:3(233).
Li, K. & Willis, R. (1992), An iterative scheduling technique for resource-constrained
project scheduling, European Journal of Operational Research, 56(3), 370–379, ISSN
03772217, doi:10.1016/0377-2217(92)90320-9.
185
Likins, G. E.; Rausche, F.; Goble, G. & Shafer, N. (1999), Pile Installation Recording
System.
Lin, Y.-C.; Cheung, W.-F. & Siao, F.-C. (2014), Developing mobile 2D barcode/RFID-
based maintenance management system, Automation in Construction, 37, 110–121,
ISSN 09265805, doi:10.1016/j.autcon.2013.10.004.
Long, L. D. & Ohsato, A. (2009), A genetic algorithm-based method for scheduling
repetitive construction projects, Automation in Construction, 18(4), 499–511, ISSN
09265805, doi:10.1016/j.autcon.2008.11.005.
Lowe, D. (1999), Object recognition from local scale-invariant features, in Proceedings of
the Seventh IEEE International Conference on Computer Vision, IEEE, ISBN 0-7695-
0164-8, pp. 1150–1157, doi:10.1109/ICCV.1999.790410.
Lowe, D. (2004), Distinctive Image Features from Scale-Invariant Keypoints, Int. J. Com-
put. Vision, 60(2), 91–110, doi:10.1023/B:VISI.0000029664.99615.94.
Lu, M.; Chen, W.; Shen, X.; Lam, H. C. & Liu, J. (2007), Positioning and tracking con-
struction vehicles in highly dense urban areas and building construction sites, Automa-
tion in Construction, 16(5), 647–656, ISSN 09265805, doi:10.1016/j.autcon.2006.11.
001.
Lu, M.; Lam, H.-C. & Dai, F. (2008), Resource-constrained critical path analysis based on
discrete event simulation and particle swarm optimization, Automation in Construction,
17(6), 670–681, doi:10.1016/j.autcon.2007.11.004.
MacCrimmon, K. R. & Ryavec, C. a. (1964), An Analytical Study of the PERT Assump-
tions, doi:10.1287/opre.12.1.16.
Makadia, A.; Patterson, A. & Daniilidis, K. (2006), Fully Automatic Registration of 3D
Point Clouds, Computer Vision and Pattern Recognition, 2006 IEEE Computer Society
Conference on, 1, 1297–1304, ISSN 1063-6919, doi:10.1109/CVPR.2006.122.
Malcolm, D. G.; Roseboom, J. H.; Clark, C. E. & Fazar, W. (1959), Application of a
Technique for Research and Development Program Evaluation, Operations Research,
7(5), 646–669, ISSN 0030-364X, doi:10.1287/opre.7.5.646.
Marler, R. T. & Arora, J. S. (2004), Survey of multi-objective optimization methods
for engineering, Structural and Multidisciplinary Optimization, 26(6), 369–395, ISSN
1615147X, doi:10.1007/s00158-003-0368-6.
Martinez, J. & Ioannou, P. (1994), General purpose simulation with Stroboscope, in
Proc. of the Winter Simulation Conference, ISBN 0-7803-2109-X, ISSN 02750708,
pp. 1159–1166, doi:10.1109/WSC.1994.717503.
Marx, A. & König, M. (2011), Preparation of Constraints for Construction Simulation,
in Proceedings of the 2011 ASCE International Workshop on Computing in Civil Engi-
neering, pp. 1–8.
186 10 References
Marzouk, M. M. & Zaher, M. M. (2015), Tracking Construction Projects Progress using

Mobile Hand-Held Devices, in Proceedings of the 5th International/11th Construction
Specialty Conference, doi:10.13140/RG.2.1.1968.7529.
Masmoudi, M. & Haït, A. (2013), Project scheduling under uncertainty using fuzzy
modelling and solving techniques, Engineering Applications of Artificial Intelligence,
26(1), 135–149, ISSN 09521976, doi:10.1016/j.engappai.2012.07.012.
Matiukas, V. & Miniotas, D. (2011), Point Cloud Merging for Complete 3D Surface Re-
construction, Electronics And Electrical Engineering, 113(7), 73–76, ISSN 2029-5731,
doi:10.5755/j01.eee.113.7.616.
Matthews, J.; Love, P. E.; Heinemann, S.; Chandler, R.; Rumsey, C. & Olatunj, O. (2015),
Real time progress management: Re-engineering processes for cloud-based BIM in
construction, Automation in Construction, 58, 38–47, ISSN 09265805, doi:10.1016/j.
autcon.2015.07.004.
Mejía, G.; Niño, K.; Montoya, C.; Sánchez, M. A.; Palacios, J. & Amodeo, L. (2016),
A Petri Net-based framework for realistic project management and scheduling: An
application in animation and videogames, Computers & Operations Research, 66, 190–
198, ISSN 03050548, doi:10.1016/j.cor.2015.08.011.
Melzner, J.; Zhang, S.; Teizer, J.; Hollermann, S. & Bargstädt, H.-j. (2012), Safety plan-
ning based on an object-oriented building model, in Proceedings of the EG-ICE Con-
ference 2012.
Merkle, D.; Middendorf, M. & Schmeck, H. (2002), Ant colony optimization for
resource-constrained project scheduling, IEEE Transactions on Evolutionary Compu-
tation, 6(4), 333–346, ISSN 1089-778X, doi:10.1109/TEVC.2002.802450.
Metropolis, N. & Ulam, S. (1949), The Monte Carlo Method, Journal of the American
Statistical Association, 44(247), 335–341, ISSN 0162-1459, doi:10.1080/01621459.
1949.10483310.
Mika, M.; Waligóra, G. & Weglarz, J. (2005), Simulated annealing and tabu search
for multi-mode resource-constrained project scheduling with positive discounted cash
flows and different payment models, European Journal of Operational Research,
164(3), 639–668, ISSN 03772217, doi:10.1016/j.ejor.2003.10.053.
Möhring, R. & Schulz, A. (2003), Solving project scheduling problems by minimum cut
computations, Management Science, 49(3), 330–350.
Naresh, B. A. L. & Jahren, C. T. (1997), Communications and tracking for construction
vehicles, Journal of Construction Engineering and Management, 123(3), 261–268, doi:
10.1061/(ASCE)0733-9364(1997)123:3(261).
Navinchandra, D.; Sriram, D. & Logcher, R. (1988), GHOST: Project network gen-
erator, Journal of Computing in Civil Engineering, ASCE, 2(3), 239 – 254, doi:
10.1061/(ASCE)0887-3801(1988)2:3(239).
187
Navon, R. (2005), Automated project performance control of construction projects, Au-

tomation in Construction, 14(4), 467–476, ISSN 09265805, doi:10.1016/j.autcon.2004.
09.006.
Navon, R. (2007), Research in automated measurement of project performance indicators,
Automation in Construction, 16(2), 176–188, ISSN 09265805, doi:10.1016/j.autcon.
2006.03.003.
Newton, I.; Motte, A. & Chittenden, N. (1850), Newton’s principia: The mathematical
principles of natural philosophy, Geo. P. Putnam.
Ng, R.; Levoy, M.; Brédif, M.; Duval, G.; Horowitz, M. & Hanrahan, P. (2005), Light
Field Photography with a Hand-Held Plenoptic Camera, Technical report, Stanford
University, doi:10.1.1.163.488.
Nguyen, A. & Le, B. (2013), 3D Point Cloud Segmentation: A survey, in Proceedings of
the IEEE International Conference on Robotics, Automation and Mechatronics, ISBN
978-1-4799-1201-8, ISSN 2158219X, pp. 225–230, doi:10.1109/RAM.2013.6758588.
Obergrießer, M.; Pfäffinger, A.; Euringer, T. & Borrmann, A. (2009), Automatisierte

Auswertung von Produktionsdaten für den geometrischen und prozessualen Soll-Ist-
Vergleich bei dem Verfahren der Pfahlherstellung, in Tagungsband des 21. Forum
Bauinformatik, Karlsruhe.
O’Connor, M.; Elkaim, G. & Parkinson, B. (1995), Kinematic GPS for closed-loop con-
trol of farm and construction vehicles, in Proceedings of the ION GPS.
Oggier, T.; Lehmann, M.; Kaufmann, R.; Schweizer, M.; Richter, M. & Peter Metzler
(2004), An all-solid-state optical range camera for 3D real-time imaging with sub-
centimeter depth resolution (SwissRanger), Proceedings of SPIE, 5249, 534–545, ISSN
0277786X, doi:10.1117/12.513307.
Ogunmakin, G.; Teizer, J. & Vela, P. (2013), Quantifying Interactions Amongst Construc-
tion Site Machines, in Proceedings of the EG-ICE Workshop on Intelligent Computing
in Engineering, Vienna, Austria.
Oloufa, A. a.; Ikeda, M. & Oda, H. (2003), Situational awareness of construction equip-
ment using GPS, wireless and web technologies, Automation in Construction, 12(6),
737–748, ISSN 09265805, doi:10.1016/S0926-5805(03)00057-8.
Özdamar, L. & Ulusoy, G. (1995), A survey on the resource-constrained project

scheduling problem, IIE Transactions, 27(5), 574–586, ISSN 0740-817X, doi:10.1080/
07408179508936773.
Panchal, J. H.; Paredis, C. J.; Allen, J. K. & Mistree, F. (2008), A value-of-information
based approach to simulation model refinement, Engineering Optimization, 40(3), 223–
251, ISSN 0305-215X, doi:10.1080/03052150701690764.
Pareto, V. (1906), Manuale di economia politica, volume 13, Societa Editrice.
188 10 References
Park, H.-S.; Thomas, S. R. & Tucker, R. L. (2005), Benchmarking of Construction Pro-

ductivity, Journal of Construction Engineering and Management, 131(7), 772–778,
ISSN 0733-9364, doi:10.1061/(ASCE)0733-9364(2005)131:7(772).
Patterson, J. H. (1976), Project Scheduling: The Impact of Instance Structure on Heuristic
Performance, Naval Research Logistics, 23(1), 95–123, doi:10.1002/nav.3800230110.
Patterson, J. H. (1984), A comparison of exact approaches for solving the multiple con-
strained resource, project scheduling problem, Management science, 30(7), 854–867,
ISSN 0025-1909, doi:10.1287/mnsc.30.7.854.
Pinson, E.; Prins, C. & Rullier, F. (1994), Using tabu search for solving the Resource-
Constrained Project Scheduling Problem, in Proceedings of the 4. International Work-
shop on Project Management and Scheduling, pp. 102–106.
Plackett, R. L. (1983), Karl Pearson and the Chi-Squared Test, International Statistical
Review, 51(1), 59–72.
Pocock, J. B.; Hyun, C. T.; Liu, L. Y. & Kim, M. K. (1996), Relationship between Project
Interaction and Performance Indicators, Journal of Construction Engineering and Man-
agement, 122(2), 165–176, ISSN 0733-9364, doi:10.1061/(ASCE)0733-9364(1996)
122:2(165).
Poirier, E. a.; Staub-French, S. & Forgues, D. (2015), Measuring the impact of BIM on
labor productivity in a small specialty contracting enterprise through action-research,
Automation in Construction, 58, 74–84, ISSN 09265805, doi:10.1016/j.autcon.2015.
07.002.
Ponz-Tienda, J. L.; Pellicer, E.; Alarcón, L. F. & Rojas-Quintero, J. S. (2015), Integrat-
ing Task Fragmentation and Earned Value Method into the Last Planner System using
Spreadsheets, in Proceedings of the Annual Conference of the International Group for
Lean Construction, pp. 63–72.
Ponz-Tienda, J. L.; Yepes, V.; Pellicer, E. & Moreno-Flores, J. (2013), The Resource Lev-
eling Problem with multiple resources using an adaptive genetic algorithm, Automation
in Construction, 29(February 2016), 161–172, ISSN 09265805, doi:10.1016/j.autcon.
2012.10.003.
Popov, V.; Mikalauskas, S.; Migilinskas, D. & Vainiunas, P. (2006), Complex usage of
4D information modelling concept for building design, estimation, sheduling and de-
termination of effective variant, Technological and economic development of economy,
12(2), 91–98, ISSN 20294913, doi:10.1080/13928619.2006.9637728.
Pradhananga, N. & Teizer, J. (2013), Automatic spatio-temporal analysis of construction
site equipment operations using GPS data, Automation in Construction, 29, 107–122,
doi:10.1016/j.autcon.2012.09.004.
Pritsker, A. A. B.; Waiters, L. J. & Wolfe, P. M. (1969), Multiproject Scheduling with
Limited Resources: A Zero-One Programming Approach, doi:10.1287/mnsc.16.1.93.
189
Rabbani, T.; Dijkman, S.; van den Heuvel, F. & Vosselman, G. (2007), An inte-
grated approach for modelling and global registration of point clouds, ISPRS Jour-
nal of Photogrammetry and Remote Sensing, 61(6), 355–370, ISSN 09242716, doi:
10.1016/j.isprsjprs.2006.09.006.
Ranganathananda, S. (1990), Swami Vivekananda and Human Excellence, Advaita
Ashrama.
Ray, S. J. & Teizer, J. (2012), Real-time construction worker posture analysis for
ergonomics training, Advanced Engineering Informatics, 26(2), 439–455, ISSN
14740346, doi:10.1016/j.aei.2012.02.011.
Riley, L. (2012), Managing and Controlling Risk in Complex Infrastructure Projects: Us-
ing Discrete Event Simulation for Stochastic Scheduling in Construction Engineering,
in Proceedings of the Autumn Simulation Multi-Conference.
Rivera, J. (2013), A Hybrid Heuristic Algorithm For Solving The Resource Constrained
Project Scheduling Problem (RCPSP)., Revista EIA, 87–100.
Rommelfanger, H. (1990), Fulpal - An Interactive Method for Solving (Multiobjective)
Fuzzy Linear Programming Problems, in Slowinski, R. & Teghem, J. (eds.), Stochastic
Versus Fuzzy Approaches to Multiobjective Mathematical Programming under Uncer-
tainty, Springer Netherlands, volume 6 of Theory and Decision Library, ISBN 978-94-
010-7449-0, pp. 279–299, doi:10.1007/978-94-009-2111-5{\_}14.
Sabuncuoglu, I. & Goren, S. (2009), Hedging production schedules against uncertainty in
manufacturing environment with a review of robustness and stability research, Interna-
tional Journal of Computer Integrated Manufacturing, 22(2), 138–157.
Sahinoglu, Z. (2008), Ultra-Wideband Positioning Systems, Technical report, Mitsubishi
Electric Research Laboratories, Massachusetts.
Saidi, K. S.; Haas, C. T. & Balli, N. a. (2002), The Value of Handheld Computers in
Construction, in Proceedings of the 19th International Symposium on Automation and
Robotics in Construction, IAARC, Washington, DC, USA.
Saidi, K. S.; Teizer, J.; Franaszek, M. & Lytle, A. M. (2011), Static and dynamic per-
formance evaluation of a commercially-available ultra wideband tracking system, Au-
tomation in Construction, 20(5), 519–530, ISSN 09265805, doi:10.1016/j.autcon.2010.
11.018.
Salem, O. & Zimmer, E. (2005), Application of lean manufacturing principles to con-
struction, Lean construction journal, 2(2), 51–54.
Salewski, F.; Schirmer, A. & Drexl, A. (1997), Project scheduling under resource
and mode identity constraints: Model, complexity, methods, and application, Eu-
ropean Journal of Operational Research, 102(1), 88–110, ISSN 03772217, doi:
10.1016/S0377-2217(96)00219-6.
190 10 References
Schäffter, M. W. (1997), Scheduling with forbidden sets, Discrete Applied Mathematics,

72(1–2), 155–166, ISSN 0166-218X, doi:10.1016/S0166-218X(96)00042-X.
Scherer, R. J. & Ismail, A. (2011), Process-Based Simulation Library for Construction
Project Planning, in Proceedings of the 2011 Winter Simulation Conference.
Schirmer, A. (2000), Case-based reasoning and improved adaptive search for project
scheduling, Naval Research Logistics, 47(3), 201–222, ISSN 0894-069X, doi:10.1002/
(SICI)1520-6750(200004)47:3<201::AID-NAV2>3.0.CO;2-L.
Schirmer, a. (2001), Resource-constrained project scheduling: An evaluation of adap-
tive control schemes for parameterized sampling heuristics, International Jour-
nal of Production Research, 39(7), 1343–1365, ISSN 0020-7543, doi:10.1080/
00207540010022368.
Schruben, L. (1983), Simulation modeling with event graphs, Communications of the
ACM, 26(11), 957–963, ISSN 00010782, doi:10.1145/182.358460.
Schruben, L. & Yücesan, E. (1993), Complexity of Simulation Models: A Graph Theo-
retic Approach, in Proceedings of 1993 Winter Simulation Conference, ISBN 0-7803-
1381-X, pp. 641–649, doi:10.1109/WSC.1993.718302.
Senouci, A. & Eldin, N. (2004), Use of genetic algorithms in resource scheduling of
construction projects, Journal of Construction Engineering and Management, 130(6),
869–877, doi:10.1061/(ASCE)0733-9364(2004)130:6(869).
Shi, B. Q.; Liang, J. & Liu, Q. (2011), Adaptive simplification of point cloud using k-
means clustering, CAD Computer Aided Design, 43(8), 910–922, ISSN 00104485, doi:
10.1016/j.cad.2011.04.001.
Shi, J. J. (1999), Activity-based construction (ABC) modeling and simulation method,
Journal of construction engineering and management, 125(5), 354–360.
Shih, N.-J. (2003), The Application of 3D Scanner in the Representation of Building
Construction Site, Technical report, NIST Special Publication SP.
Shih, N.-J. & Wang, P.-H. (2004), Point-Cloud-Based Comparison between Construc-
tion Schedule and As-Built Progress: Long-Range Three-Dimensional Laser Scanner’s
Approach, Journal of Architectural Engineering, 10(3), 98–102, ISSN 1076-0431, doi:
10.1061/(ASCE)1076-0431(2004)10:3(98).
Sigalov, K. & König, M. (2015), Similarity estimation of BIM-based schedules, in Pro-
ceedings of the EG-ICE Workshop on Intelligent Computing in Engineering, pp. 60–69.
Silpa-Anan, C. & Hartley, R. (2008), Optimised KD-trees for fast image descriptor match-
ing, Computer Vision and Pattern Recognition, 2008. CVPR 2008. IEEE Conference
on, 1–8, ISSN 1063-6919, doi:10.1109/CVPR.2008.4587638.
191
Simpson, W. P. & Patterson, J. H. (1996), A multiple-tree search procedure for the

resource-constrained project scheduling problem, European Journal of Operational
Research, 89(3), 525–542, ISSN 03772217, doi:10.1016/0377-2217(94)00247-9.
Smirnov, N. (1948), Table for Estimating the Goodness of Fit of Empirical Distributions,
The Annals of Mathematical Statistics, 19(2), 279–281, ISSN 0003-4851, doi:10.1214/
aoms/1177733256.
Son, H.; Kim, C. & Choi, K. (2010), Rapid 3D object detection and modeling using range
data from 3D range imaging camera for heavy equipment operation, Automation in
Construction, 19(7), 898–906, ISSN 09265805, doi:10.1016/j.autcon.2010.06.003.
Song, J.; Haas, C.; Caldas, C. & Liapi, K. (2005), Locating Materials on Construction Site
Using Proximity Techniques, in Proceedings of the Construction Research Congress,
doi:10.1061/40754(183)108.
Song, J.; Haas, C. T. & Caldas, C. H. (2006), Tracking the Location of Materials on
Construction Projects, Journal of Construction Engineering and Management, 132(9),
911–918, doi:10.1061/(ASCE)0733-9364(2006)132:9(911).
Špačková, O. & Straub, D. (2013), Dynamic Bayesian Network for Probabilistic Mod-
eling of Tunnel Excavation Processes, Computer-Aided Civil and Infrastructure Engi-
neering, 28(1), 1–21, ISSN 10939687, doi:10.1111/j.1467-8667.2012.00759.x.
Sprecher, A.; Kolisch, R. & Drexl, A. (1995), Semi-active, active, and non-delay sched-
ules for the resource-constrained project scheduling problem, European Journal of Op-
erational Research, 80(5), 94–102, doi:10.1016/0377-2217(93)E0294-8.
Stilla, U.; Tuttas, S.; Braun, A. & Borrmann, A. (2015), Baufortschrittskon-
trolle basierend auf photogrammetrischen Punktwolken und 4D-
Gebäudeinformationsmodellen (BIM), in Hanke, K. & Weinold, T. (eds.), Proceedings
der 18. Internationaler Geodätischen Woche, ISBN 9783879075546, pp. 157–163.
Su, Y. Y.; Hashash, Y. M. a. & Liu, L. Y. (2006), Integration of Construction As-Built
Data Via Laser Scanning with Geotechnical Monitoring of Urban Excavation, Journal
of Construction Engineering and Management, 132(12), 1234–1241, ISSN 0733-9364,
doi:10.1061/(ASCE)0733-9364(2006)132:12(1234).
Szczesny, K. & König, M. (2015), Reactive scheduling based on actual logistics data
by applying simulation-based optimization, Visualization in Engineering, 3(10), ISSN
2213-7459, doi:10.1186/s40327-015-0020-8.
Szczesny, K.; König, M.; Laußat, L. & Helmus, M. (2013), Integration of Uncertain Real-
Time Logistics Data for Reactive Scheduling using Fuzzy Set Theory, in Proceedings
of the 30th International Symposium of Automation and Robotics in Construction and
Mining (ISARC), pp. 691–698.
192 10 References
Talbot, F. B. & Patterson, J. H. (1978), An efficient integer programming algorithm with

network cuts for solving resource-constrained scheduling problems, Management Sci-
ence, 24(11), 1163–1174, ISSN 0025-1909, doi:10.1287/mnsc.24.11.1163.
Tang, P.; Huber, D.; Akinci, B.; Lipman, R. & Lytle, A. (2010), Automatic reconstruction
of as-built building information models from laser-scanned point clouds: A review of
related techniques, Automation in Construction, 19(7), 829–843, ISSN 09265805, doi:
10.1016/j.autcon.2010.06.007.
Tauscher, E. (2011), Vom Bauwerksinformationsmodell zur Terminplanung, Ph.D. thesis,
Bauhaus-Universität Weimar.
Tauscher, E.; Smarsly, K.; König, M. & Beucke, K. (2014), Automated Generation of
Construction Sequences using Building Information Models, in Proceedings of the
International Conference on Computing in Civil and Building Engineering, ISBN
4047503398, ISSN 0926-5805, pp. 745–752, doi:10.1061/9780784413616.053.
Teizer, J. (2008), 3D range imaging camera sensing for active safety in construction,
Electronic Journal of Information Technology in Construction, 13(April), 103–117,
ISSN 14036835.
Teizer, J.; Caldas, C. H. C. & Haas, C. T. C. (2007a), Real-time three-dimensional occu-
pancy grid modeling for the detection and tracking of construction resources, Journal
of Construction Engineering and Management, 133(11), 880–888, ISSN 0733-9364,
doi:10.1061/(ASCE)0733-9364(2007)133:11(880).
Teizer, J.; Lao, D. & Sofer, M. (2007b), Rapid automated monitoring of construction site
activities using ultra-wideband, in 24th International Symposium on Automation and
Robotics in Construction, ISARC 2007, 2, ISBN 9788190423519 (ISBN), pp. 23–28.
Teizer, J. & Vela, P. A. (2009), Personnel tracking on construction sites using video
cameras, Advanced Engineering Informatics, 23(4), 452–462, ISSN 14740346, doi:
10.1016/j.aei.2009.06.011.
Thesen, A. (1976), Heuristic Scheduling of Activities under Resource and Precedence Re-
strictions, Management Science, 23(4), 412–422, ISSN 0025-1909, doi:10.1287/mnsc.
23.4.412.
Thomas, P. & Salhi, S. (1998), A tabu search approach for the resource constrained project
scheduling problem, Journal of Heuristics, 4(3), 123–139.
Tocher, K. D. (1963), The Art of Simulation, The English Universities Press Ltd., doi:
10.1080/03043799408923267.
Toklu, Y. C. (2002), Application of genetic algorithms to construction scheduling with or
without resource constraints, Canadian Journal of Civil Engineering, 29(3), 421–429,
ISSN 0315-1468, doi:10.1139/l02-034.
193
Tormos, P. & Lova, A. (2001), A Competitive Heuristic Solution Technique for Resource-
Constrained Project Scheduling, Annals of Operations Research, 102(1-4), 65–81,
ISSN 02545330, doi:10.1023/A:1010997814183.
Trevor, A. J. B.; Gedikli, S.; Rusu, R. B. & Christensen, H. I. (2013), Efficient organized
point cloud segmentation with connected components, in Proceedings of Semantic Per-
ception Mapping and Exploration.
Triggs, B.; McLauchlan, P. F.; Hartley, R. I. & Fitzgibbon, A. W. (2000), Bundle
Adjustment - A Modern Synthesis, in Triggs, B.; Zisserman, A. & Szeliski, R.
(eds.), Vision Algorithms: Theory and Practice, Springer, pp. 298–372, doi:10.1007/
3-540-44480-7{\_}21.
Tuttas, S.; Braun, A.; Borrmann, A. & Stilla, U. (2014), Comparison of Photogram-
metric Point Clouds with BIM Building Elements for Construction Progress Moni-
toring, in The International Archives of Photogrammetry, Remote Sensing and Spa-
tial Information Sciences, volume XL-3, ISSN 2194-9034, pp. 341–345, doi:10.5194/
isprsarchives-XL-3-341-2014.
Tuttas, S.; Braun, A.; Borrmann, A. & Stilla, U. (2015), Validation of Bim Compo-
nents By Photogrammetric Point Clouds for Construction Site Monitoring, in ISPRS
Annals of Photogrammetry, Remote Sensing and Spatial Information Sciences, volume
II-3/W4, ISSN 2194-9050, pp. 231–237, doi:10.5194/isprsannals-II-3-W4-231-2015.
Ullman, S. (1979), The Interpretation of Structure from Motion, Proceedings of the Royal
Society B: Biological Sciences, 203, 405–426, ISSN 0962-8452, doi:10.1098/rspb.
1979.0006.
Valls, V. & Ballestín, F. (2002), A Hybrid Genetic Algorithm for the RCPSP with the
Peak Crossover Operator, in Proceedings of 8th International Workshop on Project
Management and Scheduling.
Valls, V.; Ballestín, F. & Quintanilla, S. (2004), A population-based approach to the
resource-constrained project scheduling problem, Annals of Operations Research,
131(1-4), 305–324, ISSN 02545330, doi:10.1023/B:ANOR.0000039524.09792.c9.
Valls, V.; Ballestin, F. & Quintanilla, S. (2005), Justification and RCPSP: A technique
that pays, European Journal of Operational Research, 165(2), 375–386, doi:10.1016/j.
ejor.2004.04.008.
Valls, V.; Ballestín, F. & Quintanilla, S. (2008), A hybrid genetic algorithm for the
resource-constrained project scheduling problem, European Journal of Operational
Research, 185(2), 495–508, ISSN 03772217, doi:10.1016/j.ejor.2006.12.033.
Valls, V.; Quintanilla, S. & Ballestin, F. (2001), An evolutionary approach to the resource-
constrained project scheduling problem, in Proceedings of the 4th Metaheuristics In-
ternational Conference (MIC), pp. 217–220.
194 10 References
Valls, V.; Quintanilla, S. & Ballestin, F. (2003), Resource-constrained project schedul-

ing: A critical activity reordering heuristic, European Journal of Operational Research,
149(2), 282–301, ISSN 03772217, doi:10.1016/S0377-2217(02)00768-3.
Van Peteghem, V. & Vanhoucke, M. (2014), An experimental investigation of metaheuris-
tics for the multi-mode resource-constrained project scheduling problem on new dataset
instances, European Journal of Operational Research, 235(1), 62–72, ISSN 03772217,
doi:10.1016/j.ejor.2013.10.012.
Viljamaa, E. & Peltomaa, I. (2014), Intensified construction process control using infor-
mation integration, Automation in Construction, 39, 126–133, ISSN 09265805, doi:
10.1016/j.autcon.2013.08.015.
Waligóra, G. (2009), Tabu search for discrete–continuous scheduling problems with
heuristic continuous resource allocation, European Journal of Operational Research,
193(3), 849–856, ISSN 03772217, doi:10.1016/j.ejor.2007.11.009.
Wang, X.; Truijens, M.; Hou, L.; Wang, Y. & Zhou, Y. (2014), Integrating Augmented
Reality with Building Information Modeling: Onsite construction process control-
ling for liquefied natural gas industry, Automation in Construction, 40, 96–105, ISSN
09265805, doi:10.1016/j.autcon.2013.12.003.
Webb, J. A. & Aggarwal, J. (1982), Structure from motion of rigid and jointed objects,
Artificial Intelligence, 19(1), 107–130, ISSN 00043702, doi:10.1016/0004-3702(82)
90023-6.
Weibull, W. (1951), A Statistical Distribution Function of Wide Applicability, Journal of
Applied Mechanics, 103, 293–97.
Whitehouse, G. E. (1979), GENRES: An extension of Brooks Algorithm for project
scheduling with resource constraints, Computers & Industrial Engineering, 3(3), 261–
268, doi:10.1016/0360-8352(79)90020-2.
Wiest, J. D. (1962), The Scheduling of large Projects with limited resources., Technical
report, Carnegie Institute of Technology.
Wiest, J. D. (1967), A heuristic model for scheduling large projects with limited resources,
Management Science, 13(6), B—-359.
Winstanley, G.; Chacon, M. a. & Levitt, R. E. (1993a), An integrated project planning
environment, Intelligent SystemsEngineering, 2(2), 91–106.
Winstanley, G.; Chacon, M. A. & Levitt, R. E. (1993b), Model-Based Planning: Scaled-
Up Construction Application, Journal of Computing in Civil Engineering, 7(2), 199–
217.
Wu, C. (2011), VisualSFM: A Visual Structure from Motion System.
Wu, C.; Agarwal, S.; Curless, B. & Seitz, S. M. (2011), Multicore Bundle Adjustment,
CVPR 2011.
195
Wu, I.-C. C.; Borrmann, A.; Beißert, U.; König, M. & Rank, E. (2010), Bridge construc-
tion schedule generation with pattern-based construction methods and constraint-based
simulation, Advanced Engineering Informatics, 24(4), 379–388, ISSN 14740346, doi:
10.1016/j.aei.2010.07.002.
Wübbeler, G.; Krystek, M. & Elster, C. (2008), Evaluation of measurement uncertainty
and its numerical calculation by a Monte Carlo method, Measurement Science and
Technology, 19(8), 084009, ISSN 0957-0233, doi:10.1088/0957-0233/19/8/084009.
Yagiz, S.; Gokceoglu, C.; Sezer, E. & Iplikci, S. (2009), Application of two non-linear
prediction tools to the estimation of tunnel boring machine performance, Engineering
Applications of Artificial Intelligence, 22(4-5), 808–814, ISSN 09521976, doi:10.1016/
j.engappai.2009.03.007.
Yang, J.; Arif, O.; Vela, P. A.; Teizer, J. & Shi, Z. (2010), Tracking multiple workers
on construction sites using video cameras, Advanced Engineering Informatics, 24(4),
428–434, ISSN 14740346, doi:10.1016/j.aei.2010.06.008.
Yang, J.; Vela, P. A.; Shi, Z. & Teizer, J. (2009), Probabilistic Multiple People Tracking
through Complex Situations Multiple People Tracking, in Proceedings of the 11th IEEE
International Workshop on Performance Evaluation of Tracking and Surveillance, pp.
79–86.
Yang, J.; Vela, P. a.; Teizer, J. & Shi, Z. K. (2011), Vision-based crane tracking for un-
derstanding construction activity, 2011 ASCE International Workshop on Computing in
Civil Engineering, 28(February), 258–265, ISSN 0887-3801, doi:10.1061/41182(416)
32.
Yang, Z. Y. & Wang, Z. F. (2010), Comparison between AON and AOA network dia-
grams, in Proceedings of the IEEE 17th International Conference on Industrial En-
gineering and Engineering Management, ISBN 9781424464814, pp. 1507–1509, doi:
10.1109/ICIEEM.2010.5646036.
Yates, F. (1934), Tables Involving Small Numbers and the Chi-Squared Test, Supplement
to the Journal of the Royal Statistical Society, 1(2), 217–235.
Yu, P. L. & Leitmann, G. (1974), Compromise solutions, domination structures, and
Salukvadze’s solution, Journal of Optimization Theory and Applications, 13(3), 362–
378, ISSN 00223239, doi:10.1007/BF00934871.
Zadeh, L. A. (1978), Fuzzy sets as a basis for a theory of possibility, Fuzzy sets and
systems, 1, 3–28.
Zhang, H.; Li, H. & Tam, C. (2006), Permutation-based particle swarm optimization for
resource-constrained project scheduling, Journal of computing in civil engineering, 20,
141–149.
Zhang, H.; Li, X.; Li, H. & Huang, F. (2005a), Particle swarm optimization-based
schemes for resource-constrained project scheduling, Automation in Construction,
14(3), 393–404, ISSN 09265805, doi:10.1016/j.autcon.2004.08.006.
196 10 References
Zhang, H.; Tam, C. & Li, H. (2005b), Activity object-oriented simulation strategy for
modeling construction operations, Journal of Computing in Civil engineering, 19(3),
313–322.
Zheng, Y.-J. (2015), Water wave optimization: A new nature-inspired metaheuristic,
Computers & Operations Research, 55, 1–11, ISSN 03050548, doi:10.1016/j.cor.2014.
10.008.
List of Figures 197
List of Figures
1.1 Flow chart of the optimization-simulation cycle. The process starts with
an initial plan that undergoes a simulation and optimization process. Dur-
ing the execution of the plan, data is acquired about the progress. Unless
the plan execution is done, the performance factors can require updating,
which causes another simulation and optimization run, updating the plans
respectively. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.2 Scheme of a Resource Constrained Project Scheduling Problem (RCPSP).
Rectangles represent activities, while their fill color and hatching indi-
cates the resource that is required. The hexagons represent required ma-
terials while the rounded rectangles represent resources. The parallelo-
grams represent geometric spaces. Dotted arrows indicate usages, while
solid line arrows represent precedence constraints. . . . . . . . . . . . . . 13
2.1 Flowchart of the reactive modeling and simulation cycle. The real system
is modeled as a collection of process models, which take performance
factors of involved resources into account. The performance factors are
estimated and updated by monitoring the real system. The real system
however is subject to uncertainty which in turn affects the resulting per-
formance factors. The process models are combined into a project model
(See Section 2.6), which is then simulated to generate schedules for the
real system. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.2 Shape change of beta distribution probability density function when ma-
nipulating the shape parameters α (left) and β (right). . . . . . . . . . . . 22
2.3 Change of truncated normal distribution probability density function when
manipulating the parameters µ (left) and σ (right). . . . . . . . . . . . . . 23
2.4 Change of truncated gamma distribution probability density function when
manipulating the parameters k (left) and θ (right). . . . . . . . . . . . . . 24
2.5 Change of truncated Weibull distribution probability density function when
manipulating the parameters k (left) and λ (right). . . . . . . . . . . . . . 25
2.6 Sample data of four probability distributions. Beta distribution α=10, β=3
(Top left), Normal distribution µ=0.5, σ=0.1 (Top right), Gamma distri-
bution k=7, θ=1 (Bottom left), Weibull distribution k=5, λ=1 (Bottom
right). The plots show 10, 000 samples per distribution. . . . . . . . . . . 26
198 List of Figures
2.7 Bored pile wall example illustration. The drilling rig is positioned above
the drill location (a), then the rig adjusts to be exactly upright (b), the
hole is drilled (c), and then filled with concrete (d). This illustration was
inspired by https://de.wikipedia.org/wiki/Datei:Foundation_pile_scheme.
svg . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.8 Bored pile wall example as a block diagram in the notation for the General
Purpose Systems Simulation (GPSS). The call element on the top left
outputs all the piles that need to be produced one by one. following the
arrows, the pile is then produced and is finished once the pile instance
reaches node 15, which terminates the production of that pile. . . . . . . . 32
2.9 Bored Pile Wall Example as a block diagram in the notation for the Gen-
eral Purpose Systems Simulation (GPSS) with two drilling rigs. The call
element on top outputs each pile that needs to be produced one by one. In
node 2 a drilling rig is selected for each pile, after which the respective
rig is used for production, creating the two parallel streams for seizure of
the rig. Once the drilling is seized, the streams merge again, producing
the pile, and eventually releasing the seized rig. Eventually each pile’s
production terminates in node 22. . . . . . . . . . . . . . . . . . . . . . 33
2.10 Bored pile wall example as an Activity Cycle Diagram (ACD). This dia-
gram shows the flow of the resourced as differently colored arrows. First
a hole is selected in the "next hole" idle state, going into the "DRILL"
active state, requiring the drilling rig to be in the idle state "ready". Once
drilling is done, the drilling rig can move to the idle state "done 1", while
the drilled hole can move to the idle state "done 3". Then the drilling rig
enters the "MOVE" active state, freeing the drilled hole, which goes into
the "done 4" state. The drilling rig moves into the "done 2" state, being
ready to be adjusted for the next hole. The drilled hole, can now be filled
with concrete, hence enter the "POUR CONCRETE" active state, given
the cement truck is not currently filling another hole, hence is in the "wait"
state. Once the hole is filled, the hole resource can proceed to the next hole. 34
2.11 Bored pile wall example as a CYCLONE diagram. The colors of the
arrows indicate the resource cycle, same as in Figure 2.10. The function
element "holes 1" outputs the piles that need to be produced. the combi-
state "DRILL" is then executed given the drilling rig is available. Once the
drilling is finished the drilling rig moves away in the "MOVE" active state
and adjusts in the "ADJUST" active state. The hole goes into the "holes
2" queue element, where the holes queue up, waiting for the cement truck.
The cement truck moves to the "truck" queue element when finished and
then becomes available for the combi-state "POUR CONCRETE". Once
a hole is filled with concrete, it moves to a counter element and proceeds
to the "piles done" queue element. . . . . . . . . . . . . . . . . . . . . . 36
List of Figures 199
2.12 Bored pile wall example as a multi-mode STROBOSCOPE process dia-

gram. Compared to the CYCLONE diagram in Figure 2.11, the diagram
was extended by the fork element, offering a choice between moving to
two different piles to be drilled next. For instance, the next one to the left
or to the right. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
2.13 Bored pile wall example as an activity-based construction diagram. The
diagram only consists of the four activities "DRILL", "MOVE", "AD-
JUST" and "POUR CONCRETE". The arrows indicate precedence con-
straints. The fill color of the rectangles indicate the required resource. In
order to drill, the rig needs to be adjusted, in order to move to the next
location it needs to have finished drilling, and in order to adjust it needs
to have moved to the next location. Before the concrete can be poured,
the hole needs to be drilled and the rig needs to have moved away from
the hole. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
2.14 Bored pile wall example as a non-cyclic sequential process diagram. In
this representation, each rectangle is an activity that will only be executed
once, and then marked as finished. The order of execution needs to respect
the precedence relationships indicated by the arrows and the availability
of resources, which are indicated by the fill color of the activity rectangles.
The diagram enforces the order the piles are produced in. . . . . . . . . . 39
2.15 Bored pile wall example as a non-cyclic parallel process diagram. Com-
pared to the diagram in Figure 2.14, this diagram does not enforce the or-
der of pile production and allows full parallel execution if multiple drilling
rigs or concrete trucks are present. . . . . . . . . . . . . . . . . . . . . . 40
2.16 Bored pile wall example as a non-cyclic parallel process diagram with
geometry. The diagram in Figure 2.15 extended by a representation of
geometric space, indicated by the blue parallelograms. The spaces may be
overlapping as the drilling rig might block multiple holes while positioned
above one hole. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
2.17 Bored pile wall example as an Activities-on-the-Arrows (AoA) diagram.
Each rectangle represents a state and the arrows indicate activities, caus-
ing a state change. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
2.18 Building information model and its automatically derived precedence graph.
Illustration from Braun et al. (2015a). . . . . . . . . . . . . . . . . . . . 45
2.19 Notation of simulation modeling with event graphs (Schruben, 1983).
Two states j and k are connected by an activity that requires time t and
constraints (i) to be executed. . . . . . . . . . . . . . . . . . . . . . . . . 46
2.20 Flowchart of the discrete-event simulation process. At time t = 0 the
system is initialized with all resources idling. As long as there are un-
scheduled activities the main loop continues. It creates a list of feasible
activities, selects a feasible activity, executes it and allocates the respec-
tive resources. Once no more activities are feasible, time is incremented
to the next finishing time, activities are finished and the resources freed. . 48
200 List of Figures
3.1 Bored pile wall project example illustration viewed from above. A rect-
angular pit is to be excavated. To prevent earth from caving in, two sides
of the pit have to be secured using anchored bored pile walls. . . . . . . . 51
3.2 Plot of 50 random points and their Pareto frontier illustrated by the thick
red line. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
3.3 Plot of 10 random points with the light blue star at (0.2; 0.5) indicating
the compromise point. The thin blue lines indicate the minima for each
objective, the red star in the lower left is the utopia point. . . . . . . . . . 55
3.4 Types of precedence relationships. a) Finish-Start (FS), b) Finish-Finish
(FF), c) Start-Finish (SF), d) Start-Start (SS). . . . . . . . . . . . . . . . 57
3.5 Bored pile wall project example precedence constraints. The two bored
pile walls (B1 and B2) need to be finished before the first layer of ex-
cavation (E11 and E12), after which the pile walls can be anchored (A1
and A2) followed by the second layer of excavation (E21 and E22). Fur-
thermore excavation activities E3 and E4 need to be performed. When all
excavation is finished the foundation (F) can be produced. . . . . . . . . . 58
3.6 Bored pile wall project example solution. . . . . . . . . . . . . . . . . . 59
3.7 Bored pile wall project Example after the forward pass of the critical path
method. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
3.8 Bored pile wall project example after the backward pass of the critical
path method. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
3.9 Visualization of critical path and total float times for the example in Figure
3.5. The gray bars indicate total float times, while the blue activities are
critical. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
3.10 Representation of an activity in the PERT chart notation. . . . . . . . . . 63
3.11 Anchored bored pile wall project example solution in PERT notation. The
critical path, consisting of all activities with zero float time (B2 → E12 →
A2 → E22 → F), is highlighted in blue. The corresponding Gantt Chart
is shown in Figure 3.9. . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
3.12 Bored pile wall project example with precedence and resource constraints. 65
3.13 Bored pile wall project example precedence and resource constraints. . . 66
3.14 Scheme of a Resource Constrained Project Scheduling Problem (RCPSP). 67
3.15 Bored pile wall project example resource usage. . . . . . . . . . . . . . . 68
3.16 Bored pile wall project example solution with resource constraints. . . . . 69
3.18 Bored pile wall project example solution with resource constraints. . . . . 70
3.20 Illustration of global and local left shift operations. Left shifting E3 to t1
is a local left shift (Labeled l), while shifting E4 to t3 is a global left shift
(Labeled g). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
List of Figures 201
3.21 Feasible schedule for anchored bored pile wall project example. . . . . . 72
3.22 Semi-active schedule for anchored bored pile wall project example. . . . 72
3.23 Non-delay schedule for anchored bored pile wall project example. . . . . 73
3.24 Unit time duration transformation of the schedule in Figure 3.23 . . . . . 73
3.25 Active schedule for anchored bored pile wall project example. . . . . . . 73
3.26 Priorities assigned to each activity of the anchored bored pile wall project
example. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
3.27 Schedule and resource profile created by the serial schedule generation
scheme based on the priorities assigned in Figure 3.26. The critical path
is highlighted in blue. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
3.28 Schedule and resource profile created by the parallel schedule generation
scheme based on the priorities assigned in Figure 3.26. The critical path
is highlighted in blue. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
3.29 Comparison of performance of the serial and parallel schedule generation
schemes with random problems with up to 1000 activities. . . . . . . . . 79
3.30 Flowchart of simulation concept by Beißert et al. (2010). . . . . . . . . . 81
3.31 Schedule from Figure 3.27 with sequence enforcement constraints added
by the Dori-Algorithm depicted as thick red arrows. . . . . . . . . . . . . 82
3.32 Schedule resulting from backward simulation run according to the Dori-
Algorithm based on the sequence enforcement constraints in Figure 3.31. 83
3.33 Float times and critical path resulting from Dori-Algorithm for the sched-
ule in Figure 3.31. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
3.34 Simplified bored pile wall project example with priorities on the right
inside of each activity’s box. . . . . . . . . . . . . . . . . . . . . . . . . 89
3.35 Example schedule for simplified bored pile wall project example in Figure
3.34. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
3.36 Example schedule from Figure 3.35 after swapping E1 and E2 using the
serial schedule generation scheme. . . . . . . . . . . . . . . . . . . . . . 90
3.37 Schedule from Figure 3.35 after additionally swapping B1 and B2. . . . . 90
3.38 Example schedule from Figure 3.34 after swapping B1 and B2 using the
serial schedule generation scheme. . . . . . . . . . . . . . . . . . . . . . 91
3.39 Example schedule from Figure 3.34 after swapping B1 and B2 using the
parallel schedule generation scheme. . . . . . . . . . . . . . . . . . . . . 91
3.40 Simplified bored pile wall and sheet pile wall project example with prior-
ities on the right inside of each activity’s box. . . . . . . . . . . . . . . . 92
3.41 Schedule for the example in Figure 3.40. . . . . . . . . . . . . . . . . . . 92
3.42 Schedule for the example in Figure 3.40 after swapping E1 and E2, when
using the serial schedule generation scheme. . . . . . . . . . . . . . . . . 92
202 List of Figures
3.43 Tree created from swaps for a problem with 5 activities. Each node con-
tains a vector of priorities. One pair of priorities is swapped along each
arc. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
4.1 Search Space generated by the pseudo-serial schedule generation scheme
for a two activity problem. Each axis corresponds the the priority of one
activity. The fractional parts are priority values, the integer parts are delays.100
4.2 The example problem from Section 3.4.8.8. Five activities using three
different resource types. . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
4.3 The only schedule that can be generated by the parallel schedule genera-
tion scheme for the example in Figure 4.2. . . . . . . . . . . . . . . . . 101
4.4 The optimal solution for the problem in Figure 4.2. Compared to the
solution in Figure 4.3 this is not a non-delay schedule, but requires activity
E2 to be delayed once, which is illustrated by a red double arrow. . . . . . 101
4.5 A solution for the problem in Figure 4.2 that delays activity E2 while E2
has a lower priority value than E1. This schedule cannot be generated
by either of the serial and parallel schedule generation schemes. It can
however be generated by the pseudo-serial schedule generation scheme. . 102
4.6 A solution for the problem in Figure 4.2 that delays activity E2 twice,
while E1 has a lower priority value than E2. This schedule cannot be gen-
erated by either of the serial and parallel schedule generation schemes. It
can however be generated by the pseudo-serial schedule generation scheme.102
4.7 Convergence of serial (SSGS), parallel (PSGS) and pseudo-serial sched-
ule generation schemes on the PSPLIB 30 activity benchmark set. The
Water Wave Optimization (WWO) algorithm was run using all 3 schedul-
ing schemes. In case of the pseudo-serial schedule generation scheme,
the initial population vi was varied, illustrating the transition between the
serial and parallel schedule generation schemes. . . . . . . . . . . . . . . 103
4.8 Surface of parameter values. Horizontal axes show population size np and
number of iterations i. Vertical axis shows the mean error on the PSPLIB
j30 problem set. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
4.9 Example Schedule to illustrate local optimality. Rectangles represent ac-
tivities, fill colors represent resources, and arrows represent precedence
constraints. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
4.10 The schedule from Figure 4.9 after swapping activities B and D. The swap
increases the makespan of the schedule. . . . . . . . . . . . . . . . . . . 107
4.11 The schedule from Figure 4.9 after swapping activities A and C. The swap
increases the makespan of the schedule. . . . . . . . . . . . . . . . . . . 107
4.12 The schedule from Figure 4.9 after swapping activities A and C, as well
as B and D. Applying both swaps decreases the makespan of the schedule. 108
List of Figures 203
4.13 Search tree of schedules based on swaps and delays generated using Algo-
rithm 5. The numbers in the upper right corner of each rectangle indicates
the node index. The top schedule is the initial schedule from Figure 4.9.
The schedule is changed by either a swap or a delay at each edge. Dou-
ble arrowheads indicate delays. The red rectangle indicates the optimal
solution for the problem. As the depiction contains one undirected cycle
along nodes 1,2,4, and 6, it is not strictly a tree. However, node 6 would
practically only be reached through either node 2 or 4. . . . . . . . . . . 110
4.14 The schedule from Figure 4.9 after either delaying activity B, assuming
the priority value for D is higher than that for B, or after swapping activi-
ties B and D. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
4.15 Uncertain durations of the 30 individual activities for the j30u problem
with instance 1 and parameter 1. Each activity is delayed using a beta
distributed random variable. The horizontal axes show the duration, the
vertical axes the probability. Activities are arranged from left to right,
then top to bottom. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
4.16 Schedule for the j30u problem with instance 1 and parameter 1, as shown
in Figure 4.15, using the default activity order with the parallel schedule
generation scheme. Each plot corresponds to one activity and shows the
probability of an activity being active at a certain time. The plots are
based on 100, 000 runs. Activities are arranged from left to right, then top
to bottom. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
4.17 Dependency graph of the 30 activities for the j30u problem with instance
1 and parameter 1. The numbers in parentheses are the minimum dura-
tions. Activities 1 and 32 are added source and sink activities with dura-
tion 0. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
4.18 Uncertain duration of a schedule for the j30u problem with instance 1
and parameter 1, as shown in Figure 4.15, using the default activity order
with the parallel schedule generation scheme. The vertical axis shows the
number of occurrences out of 1, 000, 000 runs. . . . . . . . . . . . . . . . 121
5.1 Matching of image features. a) and b) show two input images. The blue
dashed lines connect matching features in c) and d). . . . . . . . . . . . . 127
5.2 The location of a feature in a single two dimensional image does not re-
veal its location in three dimensional space. . . . . . . . . . . . . . . . . 128
5.3 The locations of a feature in two two dimensional images reveal its true
location, given the locations and parameters of both cameras are known. . 129
5.4 Example for two features in three images. . . . . . . . . . . . . . . . . . 129
6.1 Laser scanning an excavation pit. The laser scanner is placed in the right
of the image, illustrated by a gray square. The blue area shows the field
of view of the laser scanner. Due to the shape of the pit, the scanner can
hardly capture the entire soil surface. . . . . . . . . . . . . . . . . . . . . 135
204 List of Figures
6.2 Laser scanning an excavation pit. The laser scanner is placed in the right
of the image, illustrated by a gray square. The blue area shows the field
of view of the laser scanner. Due to the shape of the pit and an excavator
occluding parts of the area, the scanner can hardly capture the entire soil
surface. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
6.3 The operator’s perspective, looking into the excavation pit, while taking
photos for photogrammetric point cloud generation. The ten blue rect-
angles represent the photos taken by the operator. The photos need to
overlap such that they share common features. . . . . . . . . . . . . . . . 136
6.4 Taking photos of an excavation pit for photgrammetric point cloud gen-
eration. The darker area in the center is the excavation pit. Each blue
triangle shows the orientation of a photo that was taken. The operator
moves around the pit taking photos of all currently visible areas of the pit. 137
6.5 Cutting plane of excavation pit while taking photos for photogrammet-
ric point cloud generation. Each blue triangle shows the orientation of a
photo that was taken. At each location the operator takes photos of all
currently visible areas of the pit. . . . . . . . . . . . . . . . . . . . . . . 137
6.6 Camera positions of photos taken on an actual construction site. The cir-
cular area is the excavation pit and each pyramid represents a photo that
was taken. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
6.7 Point cloud generated from photographs illustrated in Figure 6.6. . . . . . 139
6.8 Raw unprocessed point cloud. Structures outside the excavation area have
been included into the point cloud. . . . . . . . . . . . . . . . . . . . . . 139
6.9 Vertical histogram of point cloud. The graph shows the number of point
on the different elevations in the point cloud. The bottom layer within the
point cloud is clearly visible as the spike at around an elevation of 30. The
surrounding ground layer can be estimated to be at the left of the graph
where the histogram flats out. . . . . . . . . . . . . . . . . . . . . . . . . 140
6.10 The point cloud from Figure 6.8 after cleaning. Points outside the exca-
vation area have been removed. . . . . . . . . . . . . . . . . . . . . . . . 141
6.11 The convex hull of an u-shaped excavation pit would have a much higher
volume than the excavation pit. The white area is the actual excavation
pit, the orange area is the convex hull. . . . . . . . . . . . . . . . . . . . 141
6.12 Convex hull of the point cloud in Figure 6.10. The hull was created using
the quickhull algorithm by Barber et al. (1996). . . . . . . . . . . . . . . 142
6.13 Meshed surface of the point cloud in Figure 6.10. The surface was gener-
ated using the Poisson surface reconstruction algorithm by Kazhdan et al.
(2006). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
6.14 Meshed surface of the point cloud in Figure 6.10. The surface was gener-
ated using the the ball pivoting algorithm by Bernardini et al. (1999). The
red circle shows a hole that was left open. . . . . . . . . . . . . . . . . . 142
List of Figures 205
6.15 The point cloud from 6.10 after all vertical duplicate points have been
removed. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
6.16 The point cloud from Figure 6.15 after refinement to fill holes. . . . . . . 144
6.17 The point cloud from Figure 6.16 after resampling to reduce the number
of points to be evenly distributed. . . . . . . . . . . . . . . . . . . . . . . 144
6.18 Overview of the system combining the presented photogrammetry ap-
proach with the video analysis system by Ogunmakin et al. (2013). . . . . 145
6.19 Dump truck state estimates for a video segment of 6 minutes duration and
the activity states in a pie chart (Bügler et al., 2016). . . . . . . . . . . . . 146
7.1 Flowchart of the overall concept of the reactive scheduling approach. An
initial schedule is built on the basis of a model of the construction project.
Once construction has started, real-time data is acquired. If construction
is not yet finished, the model is updated and it is verified whether the float
time of an activity was exceeded. If that is the case an updated schedule
is generated. This process is repeated until construction has finished. . . . 149
7.2 Example of a simple construction schedule. The excavation activity A
requires no shared resources, while the shoring activities B through E
share the same resource, of which only one instance is available. Parallel
execution of any of the activities B, C, D, and E is hence prohibited. . . . 150
7.3 The schedule from Figure 7.2 after a delay of activity A has been ob-
served. The delay propagated through the remaining schedule because
the float time of activity A has been exceeded. Activities D and E are now
scheduled to be executed later. . . . . . . . . . . . . . . . . . . . . . . . 151
7.4 The schedule from Figure 7.2 after a delay of activity A has been ob-
served, and since the float time was exceeded, a new schedule was gener-
ated. Now activity E is executed during the delay of activity A. . . . . . 151
7.5 The resource profile for the green resource of the schedule in Figure 7.2. . 152
8.1 Point cloud of the office building excavation site. . . . . . . . . . . . . . 155
8.2 Time series of point clouds of the office building excavation site. . . . . . 156
8.3 Volumes of the series of point clouds in Figure 8.2. . . . . . . . . . . . . 156
8.4 Performance factors calculated based on the volume calculations in Fig-
ure 8.3. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
8.5 Beta distribution estimated based on the performance factors in Figure
8.4 with α = 1.4427 and β = 2.7839. The mean performance was
1227.5m3 /day. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158
8.6 Point cloud of the underground parking garage excavation site. Lifting
soil out of the pit slowed down the excavation progress. . . . . . . . . . . 159
8.7 Volume plot of the underground parking garage excavation site. . . . . . . 160
206 List of Figures

ure 8.7. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
429.21m3 /day. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
8.10 Aggregate dump truck statistics for underground parking garage. The left
part shows a pie chart with the ratios of time spent in the different states.
The right part shows the loading time for the individual trucks (Bügler
et al., 2016). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
8.11 Point cloud of the hospital excavation site. . . . . . . . . . . . . . . . . . 163
8.12 Excavation progress of the hospital excavation site . . . . . . . . . . . . 164
ure 8.12. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164
488.10m3 /day. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
B.1 Deviation of best of 500, 000 random schedules from best known solu-
tions for PSPLIB problems. . . . . . . . . . . . . . . . . . . . . . . . . . 213
B.2 Deviation of algorithms from optima for the PSPLIB j30 problem set . . 213
List of Tables 207
List of Tables
2.1 p-values of χ2 Goodness-of-fit Test. . . . . . . . . . . . . . . . . . . . . 27

2.2 p-values of Kolmogorov-Smirnov goodness-of-fit test. . . . . . . . . . . 27
3.1 The 12 measures of robustness by Chtourou & Haouari (2008) . . . . . . 53
3.2 Results of the critical path method for the example in Figure 3.5. . . . . . 61
3.3 Tabular representation of the example in Figure 3.5 with added uncertainty. 62
3.4 Tabular representation of the results of applying PERT on the data from
Table 3.3. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
3.5 List of heuristic optimization procedures for the resource constrained project
scheduling problem. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
4.1 Benchmark results for j30 problem set ordered by the mean deviation
x̄ from the known optimal solutions. The column ropt lists the ratio of
optimal solutions found. nf is the number of fitness function evaluations
required per problem. t is the mean processing time. . . . . . . . . . . . 113
4.2 Benchmark results for j60 problem set ordered by the mean deviation x̄
from the best known solutions. The column rbest lists the ratio of best
known solutions found. nf is the number of fitness function evaluations
4.3 Benchmark results for j90 problem set ordered by the mean deviation x̄
from the best known solutions. The column rbest lists the ratio of best
4.4 Benchmark results for j120 problem set ordered by the mean deviation
x̄ from the best known solutions. The column rbest lists the ratio of best
4.5 The eight objective functions of the UPSPLIB problem library. . . . . . . 117
4.6 Best solutions of water wave optimization on the uncertainty-extended
PSPLIB j30u set for the different objective functions O1 to O8 . The
result measures have been obtained through 10,000 Monte Carlo runs of
the final solution. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
208 List of Tables
4.7 Mean results of water wave optimization on the uncertainty-extended

4.8 Worst solutions of water wave optimization on the uncertainty-extended
B.1 Ratio of optima of the PSPLIB j30 problems found by algorithms and the
respective mean deviations . . . . . . . . . . . . . . . . . . . . . . . . . 214
B.4 Ratio of optima of the PSPLIB j120 problems found by algorithms and
the respective mean deviations . . . . . . . . . . . . . . . . . . . . . . . 216
B.5 Best solutions of water wave optimization on the uncertainty-extended
B.6 Mean solutions of water wave optimization on the uncertainty-extended
B.7 Worst solutions of water wave optimization on the uncertainty-extended
List of Tables 209

210 A List of Symbols
A List of Symbols
α Shape parameter of the beta distribution (Section 2.2.1)

α∗ Probability of accepting a false hypothesis (Section 2.2.5)
αi Slack indicator of activity i (Section 3.2.1)
Ag The set of active activities at stage g (Section 3.4.4)
B(αj , βj ) Function taking a sample from a beta distribution with parameters αj
and βj (Section 4.5)
β Shape parameter of the beta distribution (Section 2.2.1)
γk Resource constrainedness of resource k (Section 3.4.2)
Γ Gamma function (Section 2.2.3)
C The set of already scheduled activities (Section 3.4.4)
Dg The set of delayed activities at stage g (Section 4.1)
dj Duration of activity j (Section 3.4.7)
d¯j Realized duration of activity j (Section 4.5)
dˆj Maximum increase of duration of activity j (Section 4.5)
Ei The set of possible start times for activity i (Section 3.4.7)
E(g) The set of activities which can be started precedence-feasible in stage g
(Section 3.4.4)
Er (g) The set of activities which can be started resource- and precedence-
feasible in stage g (Section 3.4.4)
e Euler’s number (Section 2.2.2)
θ Scale parameter of Gamma distribution (Section 2.2.3)
Fo The utopia point in a search space (Section 3.2.2)
Fi The finish time of activity i
F(j) Maximum finish time of all predecessors of activity j (Section 3.4.4)
J The set of all activities (Sections 3.4.2)
K The set of Monte Carlo simulation runs (Section 4.5)
κi Average percentage increase in activity i’s duration (Section 3.2.1)
k Shape parameter of the Weibull distribution (Section 2.2.4)
kr The limit of resource r (Section 3.4.7)
kir The consumption of resource r for activity i per time period (Section
3.4.7)
λ Scale parameter of the Weibull distribution (Section 2.2.4)
Li Late start time of activity i (Section 3.4.4)
µ Mean of a random variable (Section 2.2)
211
π The ratio of a circle’s circumference to its diameter (Section 2.2.2)

mk The makespan of the schedule generated in the k-th Monte Carlo simu-
lation run (Section 4.5)
N Total number of activities (Section 3.4.8.8)
Nsucci The number of direct successors of activity i (Section 3.2.1)
Oj Objective function with index j for optimizing a schedule involving un-
certainty (Section 4.5)
ξk Resource strength of resource k (Section 3.4.2)
Pj Priority value of activity j (Section 4.1)
ρik Indicator whether activity i uses resource k (Section 3.4.2)
σ Standard deviation of a random variable (Section 2.2)
Qit The set of time periods where activity i is in progress if it was started at
time t (Section 3.4.7)
Rj Fractional part of the priority value Pj of activity j (Section 4.1)
R The set of all resources (Section 3.4.7)
rik The number of units of resource k required by activity i (Section 3.4.2)
si Slack time of activity i (Section 3.3.3)
S The set of all schedules (Section 3.4.4)
SA The set of all active schedules (Section 3.4.4)
SF The set of all feasible schedules (Section 3.4.4)
SN D The set of all non-delay schedules (Section 3.4.4)
SP A The set of all pseudo-active schedules (Section 4.1)
SSA The set of all semi-active schedules (Section 3.4.4)
Θj Metric with index j for a given schedule involving uncertainty (Section
4.5)
T The set of all time periods (Section 3.4.7)
tE Expected activity duration in PERT (Section 3.3.4)
tES Early start time in PERT (Section 3.3.4)
tEF Early finish time in PERT (Section 3.3.4)
tLS Late start time in PERT (Section 3.3.4)
tLF Late finish time in PERT (Section 3.3.4)
tM Most-likely activity duration in PERT (Section 3.3.4)
tO Optimistic activity duration in PERT (Section 3.3.4)
tP Pessimistic activity duration in PERT (Section 3.3.4)
tS Slack time in PERT (Section 3.3.4)
τe Performance factor of an excavation process (Section 2.1)
Us Set of possible swaps for a schedule (Section 4.3)
Ud Set of possible delays for a schedule (Section 4.3)
uir The overall demand of resource r by activity i (Section 3.4.7)
V The set of precedence constraints (Section 3.4.7)
wi The weight of criterion i (Section 3.2.2)
φk The average of resource k required by all activities requiring at least
one unit (Section 3.4.2)
X The set of all feasible solutions (Section 3.2.2)
Ψ The total possible number of swaps (Section 3.4.8.8)
212 B Benchmark Results
B Benchmark Results
B.1 PSPLIB Results

The data listed in the tables contains the following abbreviations:
• EA: Evolutionary algorithm by Valls et al. (2001).
• GA: Genetic algorithm by Alcaraz & Maroto (2001).
• GH: Greedy heuristic algorithm by Goncharov (2011).
• HSS: Hybrid scatter search by Debels et al. (2006).
• ILDTS: Iterative limited depth tree search algorithm presented in Section 4.3.
• PSO: Particle swarm optimization algorithm by Zhang et al. (2005a).
• PSO2: Particle swarm optimization algorithm by Chen et al. (2010).
• RL: Agent-based reinforcement learning by Jedrzejowicz & Ratajczak-Ropel (2014).
• WWO: Water wave optimization algorithm by Zheng (2015), presented in Section

4.2.
B.1 PSPLIB Results 213
PSPLIB 30 PSPLIB 60
20 20
Deviation in %
Deviation in %
15 15
10 10
5 5
0 0
0 100 200 300 400 0 100 200 300 400
Problem Problem
PSPLIB 90 PSPLIB 120

20 20
Deviation in %
Deviation in %
15 15
10 10
5 5
0 0
0 100 200 300 400 0 100 200 300 400 500 600
Problem Problem
Figure B.1: Deviation of best of 500, 000 random schedules from best known solutions
for PSPLIB problems.
Limited Depth Search, d=2 Limited Depth Search, d=3

20 20
Deviation in %
Deviation in %
15 15
10 10
5 5
0 0
0 100 200 300 400 0 100 200 300 400
Problem Problem
Limited Depth Search, d=4 Random Search

20 20
Deviation in %
Deviation in %
15 15
10 10
5 5
0 0
0 100 200 300 400 0 100 200 300 400
Problem Problem
Figure B.2: Deviation of algorithms from optima for the PSPLIB j30 problem set
Algorithm Parameters ropt x̄ t nf

HSS n = 500000 99.38% 0.01% 7, 160ms 500, 000
RL s = RL123 n/a 0.02% 34, 960ms n/a
WWO np = 10000, i = 1000 97.71% 0.04% 156, 677ms 10, 061, 675
WWO np = 1000, i = 1000 96.25% 0.06% 2, 2735ms 1, 006, 492
EA 93.54% 0.10% 590ms n/a
ILDS d = 3, n = 100 93.33% 0.14% n/a n/a
WWO np = 100, i = 10000 91.04% 0.16% 14, 630ms 1, 000, 700
WWO np = 100, i = 1000 91.25% 0.17% 2, 286ms 100, 682
ILDS d = 2, n = 100 90.21% 0.21% n/a n/a
GA oc = T P _F BC n/a 0.24% n/a 5, 000
ILDS d = 5, n = 10 84.80% 0.34% n/a n/a
PSO ni = 125 96.40% 0.42% n/a 5, 000
ILDS d = 4, n = 10 80.42% 0.44% n/a n/a
WWO np = 10, i = 10000 79.58% 0.49% 2, 129ms 100, 083
ILDS d = 3, n = 10 74.38% 0.65% n/a n/a
WWO np = 1000, i = 100 70.83% 0.88% 1, 756ms 102, 087
ILDS d = 2, n = 10 67.29% 1.10% 2, 784ms 43, 402
Random n = 500000 65.42% 1.16% 49, 967ms 500, 000
WWO np = 10, i = 500 67.92% 1.23% 121ms 5, 036
WWO np = 100, i = 100 63.33% 1.52% 176ms 10, 207
WWO np = 50, i = 100 60.42% 1.83% 125ms 5, 103
ILDS d=2 53.33% 3.57% 258ms 4, 216
ILDS d=1 42.92% 5.17% 6.95ms 77
Table B.1: Ratio of optima of the PSPLIB j30 problems found by algorithms and the
respective mean deviations
B.1 PSPLIB Results 215

HSS n = 500000 93.13% 0.07% 19, 610ms 500, 000
EA 86.46% 0.17% 1, 870ms n/a
RL s = RL123 n/a 0.51% 62, 300ms n/a
GA oc = T P _F BC n/a 0.59% n/a 5, 000
ILDS d = 2, n = 100 70.83% 0.87% 330, 123ms 3, 347, 322
WWO np = 1000, i = 1000 73.75% 0.87% 69, 396ms 1, 009, 426
WWO np = 100, i = 10000 71.25% 1.29% 44, 261ms 1, 001, 044
ILDS d=3 64.38% 1.31% n/a n/a
WWO np = 100, i = 1000 70.00% 1.32% 7, 045ms 100, 940
ILDS d = 2, n = 10 59.38% 2.01% 32, 381ms 302, 723
WWO np = 10, i = 10000 64.38% 2.11% 6, 675ms 100, 119
Random n = 500000 57.30% 2.14% 218, 751ms 500, 000
GH 68.33% 2.37% n/a n/a
WWO np = 1000, i = 100 61.04% 2.78% 5, 141ms 102, 042
WWO np = 10, i = 500 52.50% 4.05% 364ms 5, 029
WWO np = 50, i = 100 53.12% 4.06% 379ms 5, 102
ILDS d=2 48.96% 4.97% 3, 096ms 30, 581
ILDS d=1 40.00% 6.72% 45ms 207

HSS n = 500000 91.04% 0.07% 38, 870ms 500, 000
EA 88.75% 0.16% 4, 900ms n/a
RL s = RL123 n/a 0.98% 77, 290ms n/a
WWO np = 1000, i = 1000 71.67% 1.61% 133, 577ms 1, 011, 275
ILDS d = 3, n = 10 65.42% 1.62% n/a n/a
WWO np = 100, i = 1000 68.12% 2.19% 14, 601ms 101, 047
ILDS d = 2, n = 10 60.83% 2.31% n/a n/a
Random n = 500000 57.71% 2.64% 353, 538ms 500, 000
WWO np = 10, i = 10000 64.38% 2.99% 13, 930ms 100, 148
WWO np = 1000, i = 100 60.42% 3.65% 10, 533ms 102, 044
WWO np = 50, i = 100 51.04% 4.92% 777ms 5, 102
WWO np = 10, i = 500 48.12% 5.27% 749ms 5, 027
ILDS d=2 47.08% 5.48% 19, 970ms 101, 220
ILDS d=1 37.71% 6.60% 179ms 369

HSS n = 500000 72.33% 0.28% 69, 570ms 500, 000
EA 50.17% 0.75% 35, 000ms n/a
GA oc = T P _F BC n/a 1.36% n/a 5, 000
RL s = RL123 n/a 2.91% 202, 410ms n/a
WWO np = 1000, i = 1000 25.33% 5.48% 259, 471ms 1, 020, 095
ILDS d=2 11.83% 6.18% 186, 940ms 541, 750
WWO np = 100, i = 1000 20.00% 7.22% 27, 658ms 101, 802
Random n = 500000 14.17% 7.41% 482, 920ms 500, 000
WWO np = 10, i = 10000 13.83% 9.05% 26, 486ms 100, 290
WWO np = 1000, i = 100 9.17% 10.52% 19, 820ms 102, 101
WWO np = 100, i = 100 6.17% 12.41% 1, 956ms 10, 209
ILDS d=1 1.67% 14.68% 866ms 994
WWO np = 10, i = 100 2.67% 14.98% 288ms 1, 021
B.2 UPSPLIB Results 217
B.2 UPSPLIB Results

XX
XX Objective
Result XXX
XX
O1 (min) 1.05 1.02 1.04 1.04 1.03 1.03 1.04 1.04
O2 (mean) 1.11 1.11 1.11 1.13 1.11 1.11 1.12 1.11
O3 (max) 1.16 1.16 1.16 1.21 1.16 1.18 1.20 1.18
O4 (stdev) 0.89 0.89 1.02 1.03 0.96 0.98 0.95 0.92
Table B.5: Best solutions of water wave optimization on the uncertainty-extended PSPLIB
XXX
X Objective
Result XXX
XX
O1 (min) 1.21 1.21 1.22 1.54 1.22 1.22 1.22 1.22
O2 (mean) 1.33 1.32 1.32 1.68 1.32 1.32 1.32 1.32
O3 (max) 1.49 1.47 1.46 1.85 1.46 1.47 1.46 1.47
O4 (stdev) 2.82 2.41 2.09 2.47 2.11 2.17 2.09 2.09
Table B.6: Mean solutions of water wave optimization on the uncertainty-extended

XXX
XXX Objective
Result XXX
O1 (min) 1.46 1.46 1.49 2.36 1.48 1.53 1.56 1.49
O2 (mean) 1.68 1.59 1.60 2.62 1.62 1.67 1.66 1.62
O3 (max) 1.91 1.97 2.11 2.77 1.88 2.09 2.14 2.06
O4 (stdev) 12.79 9.44 7.04 8.30 5.72 9.87 8.01 6.24
Table B.7: Worst solutions of water wave optimization on the uncertainty-extended

XX
XXX Objective
Result XXX
X
O1 (min) 1.09 1.08 1.09 1.09 1.09 1.09 1.08 1.09
O2 (mean) 1.16 1.15 1.15 1.15 1.16 1.15 1.16 1.16
O3 (max) 1.23 1.21 1.20 1.22 1.22 1.21 1.21 1.21
O4 (stdev) 1.48 1.57 1.48 1.56 1.51 1.56 1.54 1.58
Table B.8: Best solutions of water wave optimization on the uncertainty-extended PSPLIB
XX
XXX Objective
Result XX
XX
O1 (min) 1.27 1.27 1.27 1.48 1.27 1.27 1.28 1.28
O2 (mean) 1.38 1.37 1.38 1.61 1.37 1.38 1.38 1.38
O3 (max) 1.53 1.51 1.52 1.79 1.51 1.52 1.52 1.52
O4 (stdev) 4.27 3.79 3.70 4.36 3.68 3.63 3.56 3.56

XXX
XXX Objective
Result XXX
O1 (min) 1.62 1.55 1.62 2.03 1.61 1.61 1.64 1.64
O2 (mean) 1.81 1.73 1.80 2.28 1.75 1.79 1.84 1.80
O3 (max) 2.13 2.01 2.03 2.54 2.09 2.13 2.08 2.14
O4 (stdev) 17.38 9.85 10.16 14.18 10.44 11.50 11.11 12.47

XXX
XXX Objective
Result XXX
O1 (min) 1.13 1.11 1.10 1.19 1.11 1.12 1.13 1.13
O2 (mean) 1.21 1.16 1.16 1.27 1.21 1.22 1.18 1.20
O3 (max) 1.31 1.33 1.32 1.37 1.33 1.31 1.27 1.32
O4 (stdev) 2.06 2.05 2.15 2.50 2.28 2.13 2.00 1.65
Table B.11: Best solutions of water wave optimization on the uncertainty-extended

B.2 UPSPLIB Results 219
XXX
X Objective
Result XXX
XX
O1 (min) 1.39 1.39 1.40 1.63 1.39 1.40 1.41 1.42
O2 (mean) 1.54 1.52 1.52 1.79 1.52 1.53 1.53 1.54
O3 (max) 1.73 1.70 1.71 1.99 1.70 1.71 1.71 1.72
O4 (stdev) 6.81 6.01 5.66 6.34 5.77 5.69 5.58 5.57

XX
XXX Objective
Result XXX
X
O1 (min) 1.68 1.59 1.65 2.08 1.68 1.63 1.68 1.67
O2 (mean) 1.87 1.79 1.85 2.30 1.83 1.81 1.84 1.83
O3 (max) 2.24 2.16 2.14 2.56 2.07 2.14 2.26 2.29
O4 (stdev) 26.36 16.09 15.35 15.07 14.15 13.66 13.69 14.83


Reactive Simulation of Shoring and Excavation Processes Based On Automated Performance Monitoring Maximilian Bügler

Uploaded by

Copyright:

Available Formats

Reactive Simulation of Shoring and Excavation Processes Based On Automated Performance Monitoring Maximilian Bügler

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Reactive Simulation of Shoring and Excavation Processes Based On Automated Performance Monitoring Maximilian Bügler

Uploaded by

Copyright:

Available Formats

Ingenieurfakultät Bau Geo Umwelt

Lehrstuhl für Computergestützte Modellierung und Simulation

Reactive Simulation of Shoring and Excavation Processes based on

Vorsitzender: Univ.-Prof. Dr. rer. nat. Ernst Rank

Die Dissertation wurde am 22.06.2016 bei der Technischen Universität

2 State-of-the-Art in Modeling and Simulation of Construction Operations 18

2.6.2.5 Precedence Constraints . . . . . . . . . . . . . . . . . 43

3 State-of-the-Art in Construction Schedule Optimization 50

4 Improved Methods for Construction Scheduling 97

5 State-of-the-Art in Data Acquisition on Construction Sites 125

6 Improved Vision-based Progress Monitoring of Excavation Processes 134

7 Data-driven Reactive Scheduling 148

8 Case Studies 155

9 Summary and Outlook 166

A List of Symbols 210

B Benchmark Results 212

"Construction planning is the most crucial, knowledge-intensive, ill-structured,

1.1 Problem Statement and Scope

Figure 1.2: Scheme of a Resource Constrained Project Scheduling Problem (RCPSP).

Additionally uncertainty is an important factor in the planning process of construction

tered circumstances into account.

• Reactive Scheduling can reduce the negative consequences of unexpected delays of

• It is possible to acquire the required data (semi-)automatically from construction

1.3 Methodological Approach

The third Chapter presents the state-of-the-art in construction schedule optimization

Metaheuristic: An approximate algorithm that aims to efficiently and effectively ex-

Monte-Carlo-Simulation: The act of performing multiple simulation runs of a system

Performance Factor: A measure for the efficiency of a machine, worker, team, or a

2 State-of-the-Art in Modeling and

"Simulation is the imitation of the operation of a real-world process or sys-

2.1 Performance Factors

2.2 Modeling Uncertainty in Construction Operations

2.2.1 Beta Distribution

2.2.2 Normal Distribution

α=3, β=3 α=3, β=3

µ=2, σ=1 µ=4, σ=1

2.2.3 Gamma Distribution

k=1, θ=1 k=3, θ=1

2.2.4 Weibull Distribution

k=1, λ=1 k=1, λ=1

2.2.5 Goodness-of-fit Tests

If detailed sample data is available for an activity or a class of activities, goodness-of-fit

Beta Distribution Normal Distribution

Gamma Distribution Weibull Distribution

Table 2.1: p-values of χ2 Goodness-of-fit Test.

Table 2.2: p-values of Kolmogorov-Smirnov goodness-of-fit test.

Given a problem with an associated solution, based on complex interaction of uncertain

2.3 Modeling Costs of Construction Operations

While it is theoretically possible to implement a continuous simulation model, describing

2.5 Modeling and Simulation of Construction Processes

2.5.1 The Bored Pile Wall Example

2.5.2 General Purpose Systems Simulation (GPSS)

this reason more specialized models were developed.

2.5.3 Activity Cycle Diagram (ACD)

2.5.4 CYCLic Operations NEtwork (CYCLONE)

2.5.5 STate and ResOurce Based Simulation of COnstruction ProcEsses (STRO-

2.5.6 Activity-Based Construction (ABC)

2.6 Modeling and Simulation of Construction Projects

This representation, though, is impractical for general execution, as there could be