-
- Structuring Construction Data for use
in a Case Based Reasoning Shell for Construction Estimating
-
- Craig D. Capano, MCSM, CPC
- Western Carolina University
- Cullowhee, NC
-
- Saeed Karshenas, Ph.D., PE and
- Craig A. Struble, Ph.D.
- Marquette University
- Milwaukee, WI
-
-
Construction cost estimating is the process of
gathering all available information in order to forecast, with some degree of
accuracy, the cost of work that will be put in place on a project. The
forecast is best achieved by comparing past project information with the new
project requirements. This paper will explore the creation of case structures
as a solution to gathering and maintaining historical costs. It will also
explore the use of a Case-Base Reasoning shell, which could be utilized to
standardize collection and analysis of data. This type of system could
potentially increase the accuracy of estimating by applying heuristic and
probabilistic analysis to the captured case data. New information is used to
compute similarity and to adapt a retrieved similar case to better meet the
needs of the new problem. This “artificial intelligence” could assist in
performing the subjective and indeterminate analysis work now carried out, but
often not captured, by estimators.
-
-
Key Words:
Artificial Intelligence, Data structure, Case Based Reasoning,
Estimating
-
-
- Introduction
-
- Collection and
retrieval of historical cost data is paramount in the development of
estimates for construction work. This process is often critical because
award of contract is predicated on the lowest qualified cost to complete the
project. An estimator must make the cost forecast based on a number of
deterministic attributes of a project but also, and more difficult to
determine, a number of subjective determinations of that project. Because of
these requirements, a simple data retrieval model is not a useful tool in
querying and compiling costs. The most successful tool would be one that
could query past projects and identify costs based on the experiences that
were encountered in that particular project. In order to accomplish this it
is believed that a tool based on Artificial Intelligence, specifically Case
Based Reasoning, could be the best approach.
-
- Artificial
Intelligence (AI) can be considered as tools that emulate human thought to
help in solving problems. The academic definition of AI states that it is
the subfield of computer science that focuses on the computer’s ability to
manipulate nonnumeric symbols and infer new facts from sets of known facts (Carrico,
1989). A more practical definition is "Artificial Intelligence is the study
of ideas that enable computers to be intelligent" (Winston, 1984). The goal
of AI can be to make computers more useful. Carrico stated that the term
artificial intelligence covers a broad range of areas (Figure 1).
-
-
- Figure 1. Branches of
Artificial Intelligence (AI) (Carrico, 1989)
|
-
- The five branches
of AI are derived from cognitive science (Carrico, 1989). Cognitive science
is the field of study that seeks to understand and mimic human mental
processes. This is the science that fuels the fields of AI.
- Knowledge systems
are software systems that have structured knowledge about a field of
expertise. They are able to solve some problems within their domain by using
knowledge derived from experts in the field. This approach emphasizes data
interpretation.
- Knowledge systems
could prove useful in organizing and structuring data for construction
estimating. These systems can be applied to any kinds of knowledge in order
to solve problems within their domain. There are several approaches using
knowledge in solving problems of various domains. These include rule-based
expert systems, model-based reasoning (MBR) (Davis and Hamscher, 1986), and
case-based reasoning (CBR) (Slade, 1991).
- Rule-based expert
systems and MBR have been discussed for use in construction but have
encountered some limitations (Adeli, 1988). These limitations are largely
based on the static constraints of algorithms. Unlike rule-based expert
system and MBR, where the decision making process is encoded strictly in
explicit and static rules or algorithms, CBR systems may learn from previous
experience (by example) and can continue the learning process as they are
used and develop. CBR systems are characterized as systems that learn from
example or their environment, and may store the knowledge gained either
explicitly as
cases,
or implicitly, as a compiled or
emergent representation of their perceived environment (Aha, 2001). It is
this model which will be explored for this discussion.
-
- A CBR methodology for construction cost
estimating is proposed in this paper. Due to the high variability in type,
size, and complexity of construction projects this paper concentrates on the
use of the technique for information re-use in construction estimating for a
CMU (Concrete Masonry Unit) block wall. Object models for the CMU block wall
and its components were formed, and used to develop a prototype application
using CBR-Works, a commercially available CBR shell. The application takes
wall construction information and adds it to cases that can then be used for
case matching. Once a suitable case has been found it can be adapted to more
closely match the new project, and following this the adapted case could be
linked to an estimating database, enabling the estimator to use the
historically matched case for use with current project. It concludes by
discussing the benefits of this approach and the limitations of the system,
together with future directions.
-
-
-
Defining CBR in
Construction Cost Estimating
-
- Construction cost
estimating has traditionally been a function that requires a great deal of
experience and understanding of the building process. It has often been
referred to as both an “art” and a “science”. The science is the
quantification of the materials, equipment, and labor. The art of estimating
included the intangibles or indeterminate affects of the unknowns that are
experienced and quite prevalent, such as weather, productivity, and risk. It
is this mix of reference that makes estimating a difficult process to model
using traditional rule-based algorithm.
-
- When attempting to
define the process in an estimating system it is often difficult to account
for all of the attributing factors. Some of the indeterminate factors arise
because the project is not built in a statically controlled environment and
each project posses its own changes and unique problems. The system
determinates require a degree of subjective analysis and inferences that are
often referred to as the “art” of estimating. The ability to successfully
identify these factors and to account for their effects on the cost, often
takes years of practice and experience. The estimating “experts” draw from
past experience and cases in order to compare values to unknowns. It is
possible to learn the tools and techniques for estimating and data modeling,
but using those tools and techniques is an art that requires experience
guided by intuition (Kroenke
2000).
-
- Because cost
studies of individual projects are based on historical costs and intuitive
adjustments, it is only natural to explore how CBR could contribute to the
effort. Representing a project or a particular work package of a project, as
a multi-dimensional case, could elucidate the issues involved and the
decisions required for costing. Past experiences with real projects could be
queried to find beneficial costs which are similar, along one or more
dimensions, to old ones. The long- term goal is to make the computer a full
member of the estimating team. While this work is preliminary and the issues
to explore are numerous, the potential to integrate state-of-the-art
adaptive learning systems, while making a significant real-world
contribution, is great.
-
-
Developing the Case Structure
-
- A CBR shell was used to develop and test
case structures and retrieval queries. An evaluation copy of CBR-Works 4,
developed by Empolis, Inc., was used to build the model case for a typical
Construction Masonry Unit (CMU) wall. The evaluation copy has most of the
functionality offered in their commercial product except for the ability to
use the program through a web interface.
-
- The CMU building component is used
extensively in construction and is generally found in all types of building
projects including residential, commercial, industrial and civil projects.
For this study walls and their attributes associated with commercial
construction projects were examined. Further research could develop
attributes and cases that may be used in querying other types of
construction material assemblies.
-
- The general parameter of this case
structure is to form an intelligent query tool for matching similar walls
that have been constructed in the past on a variety of projects. The domain
chosen is estimating the costs of CMU walls. The purpose of this domain is
twofold and is intended to:
-
- 1. Represent
walls, i.e. the relevant materials and equipment used in a wall
construction.
- 2. Represent
uncertainties, i.e. the constraints and parameters that effect assembly.
-
-
Building a CBR System
-
CBR-Works
4 allows the user to build cases from a host of attributes used to model
cases. In order to develop a CBR System a number of factors and decisions
concerning the domain must be made. These can be classified in the following
functions (Technico, 2000):
 |
Case (product) definition (domain
model) – This requires the user to define the knowledge and representative
attributes related to the specific case. |
|
Initial case (product) acquisition - A
model is prepared and simulated using the software tools. |
|
Tuning the parameters of the CBR
system: cases indexing, similarity definition – This step requires an
assessment of the parameters established. It allows for adjustments of
similarity and rule interpretations. |
|
Customizing User Interface – Although
not included with the evaluation edition of CBR-Works 4, this function
allows the model to be interfaced with intelligent web applications.
|
|
Setting up an organization for case
(product) acquisition, model changes, quality control modeling – This is
the final step in the development and implementation of the model. It
relates to the functionality and use of the system and developing the work
structures and data modeling for incorporation. This function will also
not be explored with this software edition. |
-
- CBR-Works is a modeling tool with the
purpose (Technico, 2000):
|
To get insights into the techniques, difficulties and
possibilities of modeling cases. |
|
To get insights to the problem for
deciding when such an approach can be recommended. |
|
To get a first feeling in the
complexity of developing such a querying tool. |
-
- For the example, a CMU wall model may
include definition of domain and case parameters built around the following:
|
An 8” thick CMU wall is required.
|
|
However, an insulated one would be
better. |
|
Walls that were built during specific
times of year should be distinguished. |
|
Aesthetics are an important factor
(Block type, Bond, Tooling). |
-
- Size limits specifications (Long wall
requires more control joints, over 5’ requires scaffolding, ect…..)
-
- These are but a
few of the determinations that must be made in describing the wall for this
case. These are also only some of the attributes needed by the estimator
when evaluating similarity between cases.
-
-
- The Object Model
-
- In order to identify and capture the
information relevant to the cost of the CMU wall, a model must be developed
that will reflect the critical information required for comparisons. Once
the basic attributes and relationships have been determined, weighting of
these factors can be implemented. This weighting will have an affect on the
model by allowing the user to refine queries based on relevancies and the
level of uncertainties assigned.
-
- The development of the case structure for
the CMU wall must include some description of the requirements and
attributes associated with the wall. The following procedures for developing
the case were established:
For the CMU wall an example case was
defined in natural language as follows:
Wall Class Masonry Wall
This refers to a Masonry Wall that can
be built from brick, block, or stone. But each type requires some common
attributes (date built, height, length, access, productivity, crew, and
risk). This example will be based on the attributes for a CMU (Concrete
Masonry Unit) Block wall. Each block wall construction is influenced by the
block (size, type, and whether or not it is interior or exterior), bond
pattern, jointing requirements (tooling type, and mortar type), reinforcing
requirements (horizontal and vertical), number of pilasters, frequency of
corners, insulation requirements, and amount of openings). Additionally each
block wall must be defined as above or below grade. Below grade may or may
not require damproofing. Above grade may or may not require scaffolding (Schuette,
1993).
It is understood that there may be a host
of additional factors and attributes that affect the construction of the
wall. However for this example the factors will be limited to those
identified.
The description was then transposed into
a representative set of object-oriented models using attributes and
predicates. The root of Masonry_Wall was defined using the subset of classes
and relationships to define the attributes. As shown by Figure 2 the case
illustrates the components and variables needed to store cases for a CMU
wall. For this example only a CMU Block_Wall was expanded. Each attribute
was further defined as a specific type.
In order to develop the model for this
case structure, ObjectStore Database Designer was used. This tool is
generally used for the development of cases in C++, but was found to work
well in defining this case, object, related relationships and attributes, as
they will be applied to CBR-Works. This modeling technique is based on
Rumbaugh’s object modeling technique (Rumbaugh, 1991) to represent case
knowledge.
-
- Figure 2: Masonry Wall Case
|
-
- This information
can then be used as the bases for the case model in CRR-Works. The next step
is to manually transfer this information to the Concept Manager component in
CBR-Works.
-
Applying Case Structure
to CBR-Works
-
- CBR-Works 4 is a
tool that uses CBR for diagnostic support, information retrieval and product
selection. It supports the use of existing databases as knowledge sources.
However, the evaluation copy did not include the functionality of web
integration.
-
- When developing a
model to evaluate cases, a new case must be defined. The following Figure 3
is the opening screen for CBR-Works 4.0. The Concept Manager is launched and
the data is entered in the page defining the properties.
-
-
- Figure 3: CBR-Works Concept
Manager
|
-
- Next, each case
concept must be modeled in CBR-Works using attributes and relations. As can
be seen from the following example in Figure 4, each attribute can be
defined as a Name, Type, Discriminate, Mandatory, Weighted and even multiple
languages.
-
- Objects are
presented in attribute-value description. Questions defining attribute
properties are established. These questions will be posed when running the
query and they set the parameters for interaction.
-
-
- Figure 4: Defining
Attributes
|
-
- The need to define
all attributes is not necessary. The program will function with little
information. However, the more data that can be provided in the query mode,
the greater the similarity algorithms will define a solution. Even if the
input is incomplete (because as a user you may not be sure about your
demands) CBR-Works will accept your query and give you an answer using the
most successful similarity search.
-
-
-
Defining Cases and
Querying
-
- Once objects and
types have been developed and attributes assigned with related
functionalities, new cases can be created that will be used to build the
database for query. For the CMU wall example a number of cases were
developed and titled as “jobs”. These could represent all the projects in
which the company built this type of wall system.
-
- Figure 5
illustrates how the various projects could be classified based on their
object definitions. It also indicates the attributes that must be defined as
the case is being built.
-
-
-
Figure 5: CMU Block Wall Case
|
-
- In this example
the wall is being defined as one that is “Above Grade”. All attributes and
relations associated with this object are inherited from all associated
objects. Attribute definitions are required to be entered by the user. Each
attribute is associated with a pick-box that appears when the attribute is
selected. These boxes are defined when the types were developed and are
quite easy to select and enter data from. They also include the questions
that were developed in order to prompt the user for input. In the event that
the user cannot define the attribute, it will be assigned an “undefined”
value but will be stored in the case regardless.
-
-
- Consulting the Case Base
-
- Similar to the
case definition function, the user must define the attributes that are going
to be queried from the database. Figure 6 is an example query of an Above
Grade Wall.
-
-
- Figure 6: Query of Above
Grade Wall
|
-
-
The query returns the cases in an ordered
sequence based on the highest level of similarity. The similarity is
computed based on the algorithms built in the program and also the
assignment of weighting and attribute importance by the user.
-
- For this example
Job 1122 was returned from the database as having the highest degree of
similarity with a .913 rating. A 1.0 similarity rating is considered a
perfect match.
-
- The calculations are based on the
available information about the case. As can been seen by this example, both
jobs have missing or incomplete attribute definitions. The algorithm
performs a best match based on the information provided and disregards the
undefined. This could pose a problem when applying the model to a project.
The data that has been omitted may be crucial for price determination, but
the user would not have the information available when making a cost
determination.
-
-
-
Conclusion
-
- CBR-Works as a CBR
tool proved beneficial for this research. It allows a user to define
objects, relationships, attributes and types in a process. There is a great
deal of functionality that can be built into the models. The ability to
query data and cases appears quite powerful and successful in evaluating and
identifying similarities. Also, though not tried with this model, it appears
that “outside” databases can be related to the query for increased capacity
and options. The ability to access the program and databases using a
web-based interface also seems to be a good option, although not available
in the evaluation license.
-
- Current
construction estimating systems such as WinEst and Timberline lack a
systematic approach to capturing and storing past costs and particularly the
attributes that were encountered, as a complete historical database case for
future use. This means that estimates have to be produced from scratch
virtually every time a new project is to be estimated. The potential for the
use of CBR for estimating the costs of construction for a CMU block wall has
been examined and appropriate object models have been developed, along with
a database repository for project information and a prototype solution based
on a CBR software shell was presented. It has successfully demonstrated case
representation, storage, and retrieval in an adaptive system that could
retrieve similar cases for use in developing a new construction estimate.
-
- The model developed for this research is
intentionally simple and by definition limited. The types and attributes
were selected as important given the requirement of simplicity. It is
recognized however that the definition of importance will vary from
estimator to estimator. Therefore it would appear to be better from the
point of view of flexibility, if many attributes could be defined for each
case and each estimator could select those judged to be important. As
illustrated in the example, this could be done using weighting criteria, but
would need additional definition and greater interface functionality.
-
- Additional, the researcher noted that
true learning capabilities of this software, was limited. It was hoped that
there would be more functionality of the system to intuitively learn and
recommend alternates. However it was limited by the algorithm established
and did nothing to learn from mistakes. A study of additional CBR shells may
reveal greater adaptability. Further developments and study in AI such as
evolution algorithms, may also prove greater adaptability for use in
construction estimating.
-
-
-
References
-
-
Adeli, H. 1988. Expert Systems in
Construction and Structural Engineering,
Chapman & Hall
-
-
Aha, D.W., Breslow, L.A., & Muñoz-Avila, H.
2001. “Conversational Case-based Reasoning”. Applied Intelligence,
14, 9-32
-
-
Carrico, M. A., Girard, J. C. and Jones, J.
P. 1989. Building Knowledge Systems, McGraw-Hill Book Co., New York.
-
-
Davis, R., and Hamscher ,W., 1986.
"Model-based Reasoning: Troubleshooting", Exploring Artificial
Intelligence, Morgan Kaufmann publishers, Inc., pp. 239-346.
-
-
Karshenas, S., and Tse, J., 2002. "A
Case-Based Reasoning Approach to Construction Cost
-
-
Estimating" Proceedings of the International
Workshop on Information Technology in Civil Engineering, Washington,
D.C.
-
-
Kroenke, D. M., 2000. Database Processing: Fundamentals, Design and
Implementation. Upper Saddle River, N.J., Prentice Hall.
-
- Richter, M., 2001, Knowledge Containers.
ICCBR, Vancouver, BC.
-
- Rumbaugh, J., Blaha, M., Premerlani, W.,
Eddy, F., and Loresnson, W., 1991, Object-Oriented Modeling and Design,
Prentice Hall.
-
- Schuette, S. L. and Liska, R. 1994.
Building Construction Estimating, McGraw Hill.
-
- Slade, S., 1991, "Case-Based Reasoning: A
Research Paradigm", AI MAGAZINE, vol. 12,
- pp. 42-55.
-
- Tecinno, 2000. CBR:Works - Getting
Started.
-
-
Winston, P.H. 1984. Artificial Intelligence, 2nd Ed.,
Addison-Wesley publishing Co.
