Chapter 1: Application of Artificial Intelligence in Construction Management
Humankind has given itself the scientific name ‘homo sapiens’--man the wise--because our
mental capacities are so important to our everyday lives and our sense of self. The field of
artificial intelligence, or AI, attempts to understand intelligent entities as well as construct them.
These constructed intelligent entities are interesting and useful in their own right. AI has
produced many significant and impressive products so far. Although no one can predict the
future in detail, it is clear that computers with human-level intelligence (or better) would have a
huge impact on our everyday lives and on the future course of civilization (Russell & Norvig,
Definition of Artificial Intelligence
At the very beginning, let’s try to define Artificial Intelligence (AI). Definitions of AI according
to some recent textbooks are shown below:
1. “The branch of computer science that is concerned with the automation of intelligent
behavior” (Luger and Stubblefield, 1993)
2. “The study of the computations that make it possible to perceive, reason, and act'” (Winston,
3. “The art of creating machines that perform functions that require intelligence when
performed by people” (Kurzweil, 1990)
4. “The study of how to make computers do things at which, at the moment, people are better”
(Rich and Knight, 1991)
5. “A field of study that seeks to explain and emulate intelligent behavior in terms of
computational processes'' (Schalkoff, 1990)
6. “The exciting new effort to make computers think machines with minds, in the full and
literal sense” (Haugeland, 1985)
These definitions lead us to following possible conclusion
• Systems that think like humans.
• Systems that think rationally.
• Systems that act like humans
• Systems that act rationally
Artificial Intelligent makes computer act like a human. Alan Turing defined intelligent behavior
as the ability to achieve human-level performance in all cognitive tasks. To make a computer
intelligent, it would need to possess the natural language processing to enable it to communicate
successfully in human language.
The issue of acting like a human comes up primarily when AI programs have to interact with
people, as when an expert system explains how it came to its diagnosis, or a natural language
processing system has a dialogue with a user. These programs must behave according to certain
normal conventions of human interaction in order to make them understood. The underlying
representation and reasoning in such a system may or may not be based on a human model
(Russell & Norvig, 1995).
If we are going to say that a given program thinks like a human, we must have some way of
determining how humans think. Once we have a sufficiently precise theory of the mind, it
becomes possible to express the theory as a computer program. If the program's input/output and
timing behavior matches human behavior, that is evidence that some of the program's
mechanisms may also be operating in humans.
The Greek philosopher Aristotle was one of the first to attempt to codify ``right thinking,'' that is,
irrefutable reasoning processes. His famous syllogisms provided patterns for argument structures
that always gave correct conclusions given correct premises. For example, ``Socrates is a man;
all men are mortal; therefore Socrates is mortal.'' These laws of thought were supposed to govern
the operation of the mind, and initiated the field of logic. The development of formal logic in the
late nineteenth and early twentieth centuries provided a precise notation for statements about all
kinds of things in the world and the relations between them. By 1965, programs existed that
could, given enough time and memory, take a description of a problem in logical notation and
find the solution to the problem, if one exists.
Acting rationally means acting so as to achieve one's goals, given one's beliefs. An agent is just
something that perceives and acts. In this approach, AI is viewed as the study and construction of
rational agents. In this approach to AI, the whole emphasis was on correct inferences. Making
correct inferences is sometimes part of being a rational agent, because one way to act rationally
is to reason logically to the conclusion that a given action will achieve one's goals, and then to
act on that conclusion. The study of AI as rational agent design therefore has two advantages.
First, it is more general, because correct inference is only a useful mechanism for achieving
rationality, and not a necessary one. Second, it is more amenable to scientific development than
approaches based on human behavior or human thought, because the standard of rationality is
clearly defined and completely general. Human behavior, on the other hand, is well-adapted for
one specific environment and is the product, in part, of a complicated and largely unknown
evolutionary process that still may be far from achieving perfection (Russell & Norvig, 1995).
° Formal Tasks (mathematics, games)
° Mundane tasks (perception, robotics, natural language, common sense reasoning)
° Expert tasks (financial analysis, medical diagnostics, engineering, scientific analysis, and
Following is picture of task domain of AI (Rich and Knight, 1991). Artificial intelligence mainly
deals with general problem solving, theorem proving and game playing. Three domains emerges
viz. Formal task, mundane task and expert task. There are numerous branches emanates from
these three. Formal task mainly deals with mathematics problem and game playing. Mundane
task is little bit complex than formal task. Natural language processing (make English or other
languages understandable to computer), common sense reasoning (logical inferencing from
common sense), robotics and perception (sense the environment) are included in mundane task
domain. Expert tasks are the most sophisticated domain of AI. It deals with scientific analysis,
medical diagnosis, engineering, Finance analysis.
Figure 1.1: Task Domains of Artificial Intelligence
Applications of AI
Application of AI is numerous and ever increasing. Followings are most common applications.
Game Playing: One can buy machines that can play master level chess for a few hundred
dollars. There is some AI in them, but they play well against people mainly through brute force
computation--looking at hundreds of thousands of positions.
Speech Recognition: In the 1990s, computer speech recognition reached a practical level for
limited purposes. Thus United Airlines has replaced its keyboard tree for flight information by a
system using speech recognition of flight numbers and city names.
Understanding Natural Language: Just getting a sequence of words into a computer is not
enough. The computer has to be provided with an understanding of the domain the text is about,
and this is presently possible only for very limited domains.
Computer Vision: The world is composed of three-dimensional objects, but the inputs to the
human eye and computers' TV cameras are two dimensional. Some useful programs can work
solely in two dimensions, but full computer vision requires partial three-dimensional information
that is not just a set of two-dimensional views. At present there are only limited ways of
representing three-dimensional information directly, and they are not as good as what humans
Expert Systems (ES): A computer program that contains a knowledge base and a set of
algorithms or rules that infer new facts from knowledge and from incoming data. An expert
system is an artificial intelligence application that uses a knowledge base of human expertise to
aid in solving problems. The degree of problem solving is based on the quality of the data and
rules obtained from the human expert. Expert systems are designed to perform at a human expert
level. In practice, they will perform both well below and well above that of an individual expert.
The expert system derives its answers by running the knowledge base through an inference
engine, a software program that interacts with the user and processes the results from the rules
and data in the knowledge base. Expert systems are used in applications such as medical
diagnosis, equipment repair, investment analysis, financial, estate and insurance planning, route
scheduling for delivery vehicles, contract bidding, counseling for self-service customers,
production control and training.
Reasoning of Humans: Normally people create categories for reasoning. For example: cash is a
current asset and a current asset is an asset. Thus they categorize assets. They use specific rules,
a priori rules to give priority. Rules can be cascaded like:
"If A then B" . . .
"If B then C"
They also use heuristics, "rules of thumb". Heuristics can be captured using rules like “If the
meal includes red meat then choose red salad dressings". Heuristics represent conventional
wisdom. They use past experience, "cases", particularly evident in precedence-based reasoning
e.g. law or choice of accounting principles. Similarity of current case to previous cases provides
basis for action choice. Cases are stored using key attributes. Example of attributes can be shown
as: cars may be characterized by: year of car; make of car; speed of car etc. What makes good
argumentation also makes good reasoning
Computer’s Reasoning: Basically computer models are based on our models of human
reasoning. It uses frames, rules, cases and expectation.
Some AI Branches
A superset of Boolean logic dealing with the concept of partial truth -- truth values between
"completely true" and "completely false". It was introduced by Dr. Lotfi Zadeh of UCB in the
1960's as a means to model the uncertainty of natural language. Any specific theory may be
generalized from a discrete (or "crisp") form to a continuous (fuzzy) form, e.g. "fuzzy calculus",
"fuzzy differential equations" etc. Fuzzy logic replaces Boolean truth values with degrees of
truth which are very similar to probabilities except that they need not sum to one. Instead of an
assertion pred(X), meaning that X definitely has the property associated with predicate "pred",
we have a truth function truth(pred(X)) which gives the degree of truth that X has that property.
We can combine such values using the standard definitions of fuzzy logic:
truth(not x) = 1.0 - truth(x)
truth(x and y) = minimum (truth(x), truth(y))
truth(x or y) = maximum (truth(x), truth(y))
(There are other possible definitions for "and" and "or", e.g. using sum and product). If truth
values are restricted to 0 and 1 then these functions behave just like their Boolean counterparts.
This is known as the "extension principle". Just as a Boolean predicate asserts that its argument
definitely belongs to some subset of all objects, a fuzzy predicate gives the degree of truth with
which its argument belongs to a fuzzy subset.
A network of many very simple processors ("units" or "neurons"), each possibly having a (small
amount of) local memory. The units are connected by unidirectional communication channels
("connections"), which carry numeric (as opposed to symbolic) data. The units operate only on
their local data and on the inputs they receive via the connections. A neural network (NN) is a
processing device, either an algorithm, or actual hardware, whose design was inspired by the
design and functioning of animal brains and components thereof. Most neural networks have
some sort of "training" rule whereby the weights of connections are adjusted on the basis of
presented patterns. In other words, neural networks "learn" from examples, just like children
learn to recognize dogs from examples of dogs, and exhibit some structural capability for
generalization. Neurons are often elementary non-linear signal processors (in the limit they are
simple threshold discriminators).
Another feature of NNs which distinguishes them from other computing devices is a high degree
of interconnection which allows a high degree of parallelism. Further, there is no idle memory
containing data and programs, but rather each neuron is pre-programmed and continuously
active. The term "neural net" should logically, but in common usage never does, also include
biological neural networks, whose elementary structures are far more complicated than the
mathematical models used for artificial neural networks ( ANNs).
It is based on pattern recognition - used for credit assessment and fraud detection. It looks for
patterns in a set of examples and learns from those examples by adjusting the weights of the
connections to produce output patterns. Input to output pattern associations are used to classify a
new set of examples. It is able to recognize patterns even when the data is noisy, ambiguous,
distorted, or has a lot of variation. The common architecture used for ANN is feed-forward
network, shown in the next figure.
In feed forward network, an input layer with five neurons, two hidden layers with three neurons
each, and an output layer with two neurons are connected. The state function used is summation
function and the transfer functions used is sigmoid squashing function. Here training algorithm is
back-propagation algorithm. Neurons are the processing elements of network. The vocabulary in
this area is not completely consistent and different authors tend to use one of a small set of terms
for a particular concept. Neuron consists of a set of weighted input connections, a bias input, a
state function, a nonlinear transfer function, an output. The following figure shows the structure
of a neuron.
Input connections have an input value that is either received from the previous neuron or in the
case of the input layer from the outside. Bias is not connected to the other neurons in the network
and is assumed to have an input value of 1 for the summation function. Weights are real numbers
representing the strength or importance of an input connection to a neuron. Each neuron input,
including the bias, has an associated weight. The most common form of a state function is a
simple summation function. The output of the state function becomes the input for the transfer
function. A transfer function is a nonlinear mathematical function used to convert data to a
Training is the process of using examples to develop a neural network that associates the input
pattern with the correct answer. A set of examples (training set) with known outputs (targets) is
repeatedly fed into the network to "train" the network. This training process continues until the
difference between the input and output patterns for the training set reach an acceptable value.
Several algorithms used for training networks. Of them the most common is back-propagation.
Back-propagation is done is two passes: First the inputs are sent forward through the network to
produce an output, then the difference between the actual and desired outputs produces error
signals that are sent "backwards" through the network to modify the weights of the inputs.
Genetic Algorithm (GA) An evolutionary algorithm which generates each individual from some
encoded form known as a "chromosome" or "genome". Chromosomes are combined or mutated
to breed new individuals. "Crossover", the kind of recombination of chromosomes found in
sexual reproduction in nature, is often also used in GAs. Here, an offspring's chromosome is
created by joining segments chosen alternately from each of two parents' chromosomes which
are of fixed length.
GAs are useful for multidimensional optimization problems in which the chromosome can
encode the values for the different variables being optimized.
Some Application of AI in the Field of Construction
Analogy-Based Solution to Markup Estimation Problem
This paper presents a methodology for deriving analogy-based solutions to a class of
unstructured problems in civil engineering. Such problems have identifiable characteristics,
including: (1) Problems frequently require simultaneous assessment of a large number of
quantitative as well as qualitative factors that influence the solution; (2) traditional algorithmic
and reasoning-intensive techniques are not adequate to model the problem; (3) solutions are
devised in practice primarily based on analogy with previous cases coupled with a mixture of
intuition and experience; and (4) domain knowledge is mostly implicit and very difficult to be
extracted and described. For this class of problems, artificial neural networks (ANNs) are most
suited for developing decision aids with analogy-based problem-solving capabilities. A
methodology is presented and used to develop a practical model for markup estimation using
knowledge acquired from contractors in Canada and the U.S. The model design, training, and
testing are described along with the generalization improvements made using the genetic
algorithms technique. (Tarek Hegazy and Osama Moselhi, January,1994).
Neuro-modex -Neural Network System for Modular Construction Decision Making
This paper presents an approach for decision making about construction modularization using
neural networks. The model helps make a decision whether to use a conventional "stick-built"
method or to use some degree of modularization when building an industrial process plant. This
decision is based on several decision attributes which are divided into following five categories:
plant location, environmental and organizational, labor-related, plant characteristics, and project
risks. The neural network is trained using cases collected from several engineering and
construction firms and owner firms of industrial process plants. In this paper, an overview of
modular construction is provided and the reasons for using a neural network are also discussed.
The architecture, representation, and training procedure for the selected neural network
paradigms are described. The performance of the trained neural network system is compared
with the recommendations provided by human experts. The results of statistical tests performed
to validate the system are also resented (Murtaza and Fisher, 1994).
Neuroform - Neural Network System for Vertical Formwork Selection
This paper presents a neural network approach for building Neuroform, a computer system that
provides the selection of vertical formwork systems for a given building site. The reasons for
choosing a neural network approach instead of a traditional expert system are discussed. The
selection of an appropriate neural network model, its architecture, representation of the network
training examples, and the network training procedure are described. The details of the user
interaction with the trained neural network system are presented. The performance of Neuroform
is validated comparing its recommendations with that of Wallform, a rule-based expert system
for vertical formwork selection. A statistical hypothesis test, conducted on the recommendations
of Neuroform when partial inputs are given, demonstrates the system's fault-tolerant and
generalization properties. (Kamarthi, Sanvido, and Kumara, 1992).
Belief Networks for Construction Performance Diagnostics:
Belief networks, also referred to as Bayesian networks, are a form of artificial intelligence that
incorporates uncertainty through probability theory and conditional dependence. Variables are
graphically represented by nodes, whereas conditional dependence relationships between the
variables are represented by arrows. A belief network is developed by first defining the variables
in the domain and the relationships between those variables. The conditional probabilities of the
states of the variables are then determined for each combination of parent states. During
evaluation of the network, evidence may be entered at any node without concern about whether
the variable is an input or output variable. An automated approach for the improvement of the
construction operations involving the integration of the belief networks and computer simulation
is described. In this application, the belief networks provide diagnostic functionality to the
performance analysis of the construction operations. Computer simulation is used to model the
construction operations and to validate the changes to the operation recommended by the belief
network. (McCabe, AbouRizk, and Goebel, 1998).
Building KBES for Diagnosing PC Pile with Artificial Neural Network
Diagnosis of damage of prestressed concrete piles during driving is an important problem in
foundation engineering. An effort to build an expert system for the problem is described in this
paper. To overcome the bottleneck of knowledge acquisition, an artificial neural network is used
as the learning mechanism to transfer engineering experience into usable knowledge. The back-
propagation learning algorithm is employed to train the network for extracting knowledge from
training examples. The influences of various control parameters (including learning rate and
momentum factor) and various network architecture factors (including the number of hidden
units and the number of hidden layers) are examined. The results prove that the artificial neural
network can work sufficiently as a knowledge-acquisition tool for the diagnosis problem. To
apply the knowledge in the trained network, a reasoning strategy that hybridizes forward-and
backward-reasoning schemes is proposed to realize the inference mechanism. (Yeh, Kuo and
Modeling Initial Design Process using Artificial Neural Networks
The preliminary design model is of vital importance in the synthesis of a finally acceptable
solution is a design problem. The initial design process is extremely difficult to computerize
because it requires human intuition. It has often been impossible to form declarative rules to
express human intuition and past experience. The suitability of an artificial neural network for
modeling an initial design process has been investigated in this paper. Development of a network
for initial design of reinforced-concrete rectangular single-span beams has been reported. The
network predicts a good initial design (i.e., tensile reinforcement required, depth of beam, width,
cost per meter, and the moment capacity) for a given set of input parameters (i.e., span, dead
load, live load, concrete grade, and steel type). Various stages of development and performance
evaluation with respect to rate of learning, fault tolerance, and generalization have been
presented. (Mukherjee and Deshpande, 1995).
Intelligent Planning of Construction Projects
Knowledge representation and reasoning techniques derived from artificial intelligence (AI)
research permit computers to generate plans, not merely analyze plans produced by humans.
They explicitly represent knowledge about how to generate plans in the form of initial and goal
states, descriptions of actions along with their preconditions and effects, and a control structure
for selection new actions to insert into a project plan. From the more than two dozen AI planners
developed and published since the 1960s, we have chosen the system for interactive planning
and execution (SIPE) to investigate the utility of AI planners for construction project planning.
This paper presents our experience modeling a multistory office building project for construction
planning, implementing SIPE to plan this project, and describing SIPE's performance in planning
the construction of large-scale multistory buildings. With the use of a frame hierarchy, generic
operators, and a constraint-based approach, SIPE can generate logically correct activity networks
for multistory building construction from a description of the components of a facility. To model
such construction projects in a concise and uniform framework, we show the usefulness of some
underlying principles for establishing ordering relationships among the project components
involved in construction activities.m( Kartam, and Levitt, 1990)
Construction Robot Fleet Management System Prototype
The application of robotic equipment to the execution of construction tasks is gaining attention
by researchers and practitioners around the world. A number of working prototype systems have
been developed by construction companies or system manufacturers, and implemented on
construction job sites. Several Japanese construction firms have already developed their own
fleet of construction robots. This paper describes a HyperCard (trademark) prototype of the
construction robotic equipment management system (CREMS), developed as a response to the
need to effectively manage diverse robots on future construction sites. Modules comprising the
system and their initial implementation within the CREMS prototype are presented. An example
consultation with the system is provided. Ongoing system developments and embellishments are
also outlined. The utility of this system lies in optimizing the robot performance of work tasks on
as many construction projects in a contractor's portfolio as feasible. Thus, economic benefits of
robot use can be achieved more easily. Thus, robot development costs can be recovered faster,
and robot use can be distributed over more applications and types of construction tasks
(Skibniewski, and Russell, 1991).
Bridge Planning Using GIS and Expert System Approach
In the planning process of a new road network, the planner should consider possible locations of
bridges and tunnels. The selection of the best alignment imposes the need to investigate the
effect of the location of each bridge on the bridge type that fits this location. This task has not
been done so far because of the large volume of data needed and the complicated interaction
between many factors. In this paper, it is shown that considering this task in the early stage of
road alignment planning can result in a more rational design. Geographic information systems
and expert systems are proposed as two methodologies that can help in comparing candidate sites
and candidate types simultaneously. Having this computation power, quantitative comparison
can be done faster and much more precisely than in the case of conventional simplified methods.
This can result in improving the design of the road network in general and in having bridges
designed to meet the requirements of erection, maintenance, driving comfort, and landscape.
(Hammad, Itoh, and Nishido, 1993).
Comparison of Case-Based Reasoning and Artificial Neural Networks
The outcome of construction litigation depends on a large number of factors. To predict the
outcome of such litigation is difficult because of the complex interrelationships between these
many factors. Two attempts are reported in the literature that use, respectively, case-based
reasoning (CBR) and artificial neural court cases; and additional 12 cases were used for testing.
Prediction rates of 83% in the CBR study and 67% in the ANN study were obtained. In this
paper, CBR and ANN are compared, and their advantages and disadvantages are discussed in
light of these two studies. It appears that CBR is more flexible when the system is updated with
new cases, has better explanation facilities, and handles missing data and a large number of
features better than ANN in this domain. If the use of CBR and ANN is understood better and if,
as a result, the outcome of construction litigation can be predicted with reasonable accuracy and
reliability, all parties involved in the construction process could save considerable money and
time. (Arditi and Tokdemir, 1999 ).
Site-Level Facilities Layout Using Genetic Algorithms
Construction site-level facilities layout is an important activity in site planning, the objective of
this activity is to allocate appropriate locations and areas for temporary site-level facilities such