G E O M E T R I C - B A S E D R E A S O N I N G S Y S T E M FOR

PROJECT PLANNING
By A. A. Morad, ~ Associate Member, ASCE,
and Y. J. Beliveau,2 Member, ASCE

ABSTRACT: Traditional planning and scheduling techniques provide construction

planners with mathematical models to simulate the construction process as an aid
Downloaded from ascelibrary.org by University of Exeter on 07/14/15. Copyright ASCE. For personal use only; all rights reserved.

in planning and control of complex projects. These techniques possess many lim-
itations. They usually place considerable reliance on planners' judgment, imagi-
nation, and intuition. They require abstract visualization of the perceived config-
urations, characteristics, and spatial relationships among various components of
the designed facility, which usually results in differences of opinion among various
planners. There is a need for the development of new techniques to augment and
overcome some of the limitations that exist in current planning techniques. This
paper presents a geometric-based reasoning system called KNOW-PLAN. The
system integrates artificial intelligence with 3-D computer modeling technologies
to automatically generate and visually simulate project plans. The system utilizes
mainly the geometric data that defines the topography of the designed facility to
generate the project plan.

INTRODUCTION

Planning and scheduling are vital elements to the successful execution of

any project. Clough and Sears (1979) defined planning as "the devising of
a workable scheme of operations to accomplish an established objective
when put into action." The planning aspect of any project consists of iden-
tifying the work activities n e e d e d to achieve project completion and to
sequence them, based on different project constraints and requirements, to
obtain peak performance.
Traditional planning and scheduling techniques provide construction plan-
ners with mathematical models to simulate the construction process as an
aid in planning and control of complex projects. They possess many limi-
tations. They are able to manipulate only the d a t a provided by the planners
during the planning process and not the knowledge used to generate p r o j e c t
plans. These techniques place considerable reliance on planners' j u d g m e n t ,
imagination, and intuition. These planning techniques require abstract vis-
ualization of the perceived configurations, characteristics, and spatial re-
lationships among various components of the project. This perception usu-
ally results in differences of opinion among various planners.
There is a need for the d e v e l o p m e n t of new techniques to augment and
overcome the limitations in current planning tools. These new techniques
rely on advanced visualization tools in addition to experts' knowledge to
enhance and automate the planning process of construction projects. T h e
availability of advanced c o m p u t e r technologies, both in hardware and soft-
ware, provides the capabilities for researchers and practitioners in the con-
struction industry to develop such techniques.
This p a p e r presents the implementation of a geometric-based reasoning

~Asst. Prof., Dept. of Constr. Mgmt., Florida Int. Univ., Miami, FL 33199.
-~Assoc. Prof., Dept. of Cir. Engrg., Virginia Tech., Blacksburg, VA 24061.
Note. Discussion open until June 1, 1994. To extend the closing date one month,
a written request must be filed with the ASCE Manager of Journals. The manuscript
for this paper was submitted for review and possible publication on July 30, 1992.
This paper is part of the Journal of Computing in Civil Engineering, Vol. 8, No. 1,
January, 1994. 9 ISSN 0887-3801/94/0001-0052/$1.00 + $. 15 per page. Paper
No. 4510.
52

J. Comput. Civ. Eng. 1994.8:52-71.

 project planning system called KNOW-PLAN. Geometric-based reasoning
is the process of automating the generation of project plans, for the designed
facility, mainly based on the spatial relationships that exist among various
components of the facility.

3-D COMPUTER MODELING AND VISUAL SIMULATION FOR

PROJECT PLANNING
Downloaded from ascelibrary.org by University of Exeter on 07/14/15. Copyright ASCE. For personal use only; all rights reserved.

Most recent computer-aided design (CAD) systems possess the capability

to define objects in three dimensions (3-D), 3-D wire-frame modeling and
3-D solid modeling are two object representation approaches in 3-D com-
puter modeling used for designing and creating the collection of data that
describes the configuration of a designed facility. 3-D computer models can
be used for the following tasks (Encarnacao et al. 1986):

1. Determination of geometric forms.

2. Proof of geometric compatibility.
3. Determination of physical properties.
4. Representation of the model with specific visualization techniques.

The potential of CAD systems in construction presents a significant op-

portunity for integrating engineering and construction processes in a more
cost-effective way. The combination of computer graphics, animation, and
3-D computer modeling has proven to be an extremely effective tool for
real-time animation in support of engineering and construction from the
conceptual design to the construction process.
CAD systems are widely used as planning tools in the manufacturing
industry. The construction industry, meanwhile, is still not utilizing these
systems effectively for planning and scheduling of construction projects
(Morad and Beliveau 1991). Technically advanced construction and 3-D
computer modeling systems can provide powerful graphical computer sup-
port for construction planning. This support is concerned with the pictorial
representation of the project components and handling equipment. In ad-
dition, these systems can provide visual simulation capabilities to model the
construction process graphically. This allows planners to develop and re-
hearse project plans very quickly and graphically demonstrate, in real time,
those plans to others.
There are many active groups of practitioners and researchers in the
construction industry who are involved in the development of new planning
techniques using CAD and 3-D computer modeling systems. The research
and development (R and D) divisions of many construction and engineering
firms such as Bechtel, Hitachi, and Stone & Webster are among these active
groups.

ARTIFICIAL INTELLIGENCE FOR PROJECT PLANNING

Artificial intelligence (AI) provides the potential to enable computers to
emulate functions carried out by humans. AI technology explicitly attempts
to move the reasoning process into the program. Knowledge-based expert
systems (KBESs) are examples of AI technology. They use logical rela-
tionships to embody knowledge about a specific domain and perform spe-
cialized tasks that typically require human judgment and expertise.
KBESs provide effective tools for project planning. The heuristic and
53

J. Comput. Civ. Eng. 1994.8:52-71.

 procedural knowledge needed to solve the project planning problem can be
easily stored and manipulated in the (IF [conditions] THEN [actions]) form.
They also provide reasoning tools to automatically generate project plans
that cannot be produced directly by traditional planning techniques.
The development of KBESs for planning and scheduling has been initiated
by many researchers and practitioners. There are many completed and
ongoing research efforts aimed at the utilization of KBESs in construction
Downloaded from ascelibrary.org by University of Exeter on 07/14/15. Copyright ASCE. For personal use only; all rights reserved.

planning to perform various automated planning functions. Refer to Dym

and Levitt (1991) for a list of some of the early implemented AI planners
and ongoing research in the area of project planning.

KNOW-PLAN APPROACH
Advancement in computer technology such as 3-D computer modeling
and artificial intelligence (AI) are offering new and potentially powerful
capabilities to develop innovative techniques to enhance and automate the
project planning process. The integration of such technologies to generate
and visually simulate construction plans provides a radical departure from
the conventional planning approach. The integration effort can be based
on common representation of the project components with attributes that
are passed from a 3-D computer modeling system to a geometric-based
reasoning system.
Construction planners are professional experts who are knowledgeable
about the process of sequencing the different activities of construction proj-
ects. This process is based on planners' previous experience, intuition, and
perception of the configuration and topology of the facility to be constructed.
There are many issues considered during the typical planning process. Some
of these issues are: spatial relationships among various components of the
facility, design constraints, constructability constraints, resource constraints,
the plan's level of detail, method of construction, weather constraints, laws
of nature, stability constraints, laws and regulations, and project specifi-
cations. However, the main source of knowledge needed to define the
project plan is the perception of the spatial relationships among various
components of the facility.
The KNOW-PLAN system is a geometric-based reasoning system for
project planning that utilizes geometric data to provide a dynamic sequenc-
ing process for project planning. The system utilizes component location,
interaction, and spatial relationships with other components as the primary
source of reasoning for the project plan. The interaction of components is
based on classification of components with relation to connection types, the
zones in which they are located, and relationships between the classes with
which they are associated.
The spatial relationships among various components can be extracted
from the 3-D computer model of the designed facility. This information
with other knowledge captured from construction planners can be stored in
the knowledge base of the system. The stored knowledge is used in the
reasoning process to generate automatically the project plan. Using an ad-
vanced visual simulation system, the construction process can be visually
animated using the generated plan and the 3-D computer model of the
facility. The animation helps in manually identifying any conflict in the
generated plan which is not properly resolved by the limited knowledge
available in the knowledge base of the system.
54

J. Comput. Civ. Eng. 1994.8:52-71.

 KNOW-PLAN SYSTEM DESCRIPTION AND ARCHITECTURE
In order to prove the concept of utilizing geometric data and spatial
relationships among the project components in project planning, an overall
model called the KNOW-PLAN model has been formulated (Morad and
Beliveau 1991). This conceptual model demonstrates theoretically the fea-
sibility of implementing such a model. The actual implementation effort
concentrates on the development of the crucial components of the KNOW-
Downloaded from ascelibrary.org by University of Exeter on 07/14/15. Copyright ASCE. For personal use only; all rights reserved.

PLAN model using advanced computer techniques. In this paper, the actual
implementation of the KNOW-PLAN system is presented.
The KNOW-PLAN system consists of four modules:

1. 3-D computer modeling module.

2. Knowledge and data module.
3. Dynamic sequencing module.
4. Visual simulation model.

3-D COMPUTER MODELING MODULE

In order to utilize CAD systems in the development of an advanced
planning system, the planning system should interact with a 3-D computer
modeling system to use existing component definition data. The components
of a 3-D computer model of a designed facility represent the physical objects
that are associated with the different activities or group of activities in
construction plans. The interaction can be accomplished by developing a
communication environment for data exchange, which means that compo-
nent definition data has to be extracted and transferred to a format readable
by the planning system. The KNOW-PLAN system utilizes W A L K T H R U ,
a r e a l - t i m e 3-D visual s i m u l a t i o n s y s t e m d e v e l o p e d by B e c h t e l
( " W A L K T H R U " 1991), to provide the necessary geometric data needed
during the dynamic sequencing process.
The purpose of the 3-D computer modeling module is to generate a 3-D
computer model, conduct the interface with W A L K T H R U , and perform
the data extraction process. These tasks and data flow among them are
illustrated in the left half of Fig. 1. The following is a detailed discussion
of the execution of these tasks.
A 3-D computer model of the designed facility should be generated using
a CAD system that can interface with the W A L K T H R U system. The 3-D
computer model contains the geometric definitions of the components of
the designed facility. In KNOW-PLAN, each component in the 3-D com-
puter model is associated with one activity in the project plan. Future ex-
tensions will provide the capability to associate many components with one
activity or one component with many activities at different levels of plan
abstraction with hierarchical component and activity structure definitions.
The second task in this module is to translate the 3-D computer model
into W A L K T H R U . Several interface programs currently exist to transfer
3-D computer models into W A L K T H R U model files. When a 3-D computer
model is converted to run with W A L K T H R U , the model can be subdivided
into a number of parts, referred to in W A L K T H R U as "objects."
When a W A L K T H R U conversion program is run, CAD design files
become objects in the W A L K T H R U model file. In this manner, a
W A L K T H R U model file can be constructed from several IGDs, 3DM, or
IGES files. Alternatively, a single large IGDS, 3DM, or IGES file can be
converted into multiple W A L K T H R U objects by running the conversion
55

J. Comput. Civ. Eng. 1994.8:52-71.

 Downloaded from ascelibrary.org by University of Exeter on 07/14/15. Copyright ASCE. For personal use only; all rights reserved.

!
3D Computer Modeling Module < ~ I <:~ Knowledge and Data Entry Module
D [ " l l ~ m m User Interaction
User ~,interaction
S WALKIGBS
~risual Simulation DATABASE
I [ System
'N
O WALKIGDS "~1 Activity'Data
III
..... ) - - ~ - -
T 3s I
I Cla~ Data I- ~ - -

WALK3DM
[Dominatin~Class Relations I -- ~ - -
[ Special Class Relations [ - - ~ _

ol
o -4~-- illll
03 W,~I~PRE
D A P~oc~ [Act. Connection Relations ] - - ~ -- R
E- ~ - 4 ~-- C [- Zoning Data I -- ~ - - T
L I ~ [ W~KaDS
i ' ~ - - " ' ~ the Central Database)

J. Comput. Civ. Eng. 1994.8:52-71.

~ - ~ ~-- A I
CADAN
Interface c I User ~ User Defined Knowledge
T I )
I Constructabilit~ Constraints )
I

User Interaction I External Knowledge Sources)

I
I
3D Computer Modeling Module ~ ~< ~ Knowledge and Data Entry Module
FIG. 1. 3-D Computer Modeling and Knowledge and Data-Entry Modules
 program multiple times, repeatedly specifying the same input and output
files, while using one or more options to limit the entities processed with
each pass. In this manner, each run of the conversion program creates a
single object in the W A L K T H R U model file from the parts specified by
the large IGDS, 3DM, or IGES file ( " W A L K T H R U " 1991). Each object
in a W A L K T H R U model file corresponds to one component in the KNOW-
PLAN system.
Downloaded from ascelibrary.org by University of Exeter on 07/14/15. Copyright ASCE. For personal use only; all rights reserved.

The KNOW-PLAN system extracts the geometric data needed for the
reasoning and the visual simulation processes within W A L K T H R U from
two sources. The first source is a volumefile that defines the maximum and
minimum values of the objects' boundaries in the x-, y-, and z-directions.
The file is a simple ASCII file that lists the needed values for each object
in the 3-D computer model. Another alternative for getting the geometric
data is during the model conversion process using a volume definition file
that relates to the model file. Refer to the W A L K T H R U manual regarding
the structure of this file.
The second source for the data to be extracted is an initial record file.
Record files in W A L K T H R U contain a sequence of one or more key frames.
Each key frame contains view configuration and optional object orientation
data about the model at a particular moment in time. The initial record file
is a file that contains only one key frame for which all the objects of the
model are displayed. The file will contain the coordinates of the centers of
the objects and the rotations of the objects in the x-, y-, and z-directions.
The geometric data extracted from the initial record file and the boundary
data extracted from the volume definition file of the model are transferred
to the system's central database for later processing.
A sample culvert project that consists of 23 components is shown in Fig.
2(a). The 3-D computer model of the project was created using the C A D A M
system. This project will be used throughout the paper as an example.

KNOWLEDGE AND DATA ENTRY MODULE

When many applications need to be integrated, an efficient data interface
environment should be developed. In KNOW-PLAN, geometric data for
the 3-D computer model needs to be passed to the dynamic sequencer. This
data will be used in the reasoning process, with other construction knowl-
edge, to generate a project plan. In order to visually simulate the construc-
tion process, the schedule attributes of the activities, based on the generated
plan, should be passed to the 3-D computer modeling system.
The dynamic sequencer needs information regarding the classes to which
components are related, the relationship between classes, and the connec-
tion type between different components. This information must be entered
by the user, or read from files of previous projects.
A central database-management system is developed as part of the KNOW-
PLAN system to perform the aforementioned tasks of data exchange, def-
inition, a n d manipulation. All the data and knowledge required by the
dynamic sequencer are entered into the system via the central database-
management system. The central database has the functionality of allowing
the user to manipulate all the data stored in the database files and to read
and write ASCII files. The following is a description of the different data
that resides in the central database. The right half of Fig. 1 serves as an
illustration of the knowledge and data-entry module.

57

J. Comput. Civ. Eng. 1994.8:52-71.

Downloaded from ascelibrary.org by University of Exeter on 07/14/15. Copyright ASCE. For personal use only; all rights reserved.

(a) 3D View o f the Sample Culvert Project (b) Zoning Data

(c) Activity Dam

:::::::::::::::::::::::::::::::::::::::::::::::::::::::::::: (d) CO~X2(PCtiO,~,Z~lTM

i!:.~::!i::::~z:::::: ~ :~t~ ::~f ::::. !i~c~i il :::. i .....................
ii~ii~ ii::::: i ! i i:i I i~ : i:i :i=i :]~]i;iil
:C1 : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : :

Iii:

(c) Class Dam (f) Dominating Cla-~s

FIG. 2. Sample Culvert Project

Activity Data
The primary objective of the KNOW-PLAN system is to plan the se-
quence of executing the project's activities. The sequence is defined by
asserting precedence relationships among different activities in the plan. In
KNOW-PLAN, the activity list of a project is defined according to the
breakdown of the project into components as defined by the 3-D computer
model of the facility. The description of an activity consists of three groups
of attributes. The first group describes the activity name, and number. The
second group comprises of the geometric data of the physical component
58

J. Comput. Civ. Eng. 1994.8:52-71.

 associated with the activity. The third group is the schedule attributes, which
include activity duration and the early and late start and finish dates of the
activity. The activity duration is input by the user. The early and late start
and finish dates of the activity are calculated by the dynamic sequencer
based on the project plan generated by the reasoning process.

Class Data and Class Relationships

Downloaded from ascelibrary.org by University of Exeter on 07/14/15. Copyright ASCE. For personal use only; all rights reserved.

A class is an assignment of an object that is related in behavior with

similar objects. The attributes of a class entity are the basic elements of
geometric knowledge in KNOW-PLAN. This knowledge is used by the
system to sequence activities, based on the spatial relationships among the
components associated with these activities. A class can define trade, design
elements, group of activities that need similar resources, or group of activ-
ities that are constrained by specific requirements. Classes can be broken
down into different subclasses, thus defining a hierarchical structure.
The KNOW-PLAN system provides many attributes to define the knowl-
edge about a specific class. A typical class has the following attributes: name,
direction of installation, and priority of installation. The direction of in-
stallation represents the knowledge, based on experts' experience, about
the sequencing of the activities that are associated with the class in the x-,
y-, and z-directions. The priority of installation has two variables: first
priority and second priority. The first priority defines which direction of
installation is dominant. The second priority defines the second level of
dominance.
The aforementioned discussion is only valid if the two activities to be
sequenced are of the same class. In the case of different classes, class re-
lationships are available in the system. Two types of relationships between
different classes are defined in KNOW-PLAN. Dominating class relation-
ships define which class among two classes is dominating with a specific type
of connection. If class C1 is dominating class C2 in the case of a (tl) type
of connection, the activity related to class C1 will be sequenced before the
activity related to class C2. This assumes that the two physical components
associated with the two activities are connected with a tl type of connection.
The second relationship type is the special class relationships. These rela-
tionships specify the situations in which a dominating class relationship is
not applicable.

Activity-Class Relationships
The user has to associate each activity to one or more classes in order to
utilize the knowledge defined about different classes in the geometric rea-
soning process. Relating activities to classes allows the activities to inherit
the direction of installation data from the related classes. Two activities of
different classes can inherit the dominating and special class relationships
knowledge of their associated classes. This knowledge is used in the geo-
metric reasoning process to come up with a sequence of installation for the
two components associated with these two activities.

Activity Connection Relationships

In construction projects, physical components are related by the type of
connection that exist among them. A component can be related to another
component by one or more means of connections. Typical connection types
are: structurally supported, embedded in, protected by, and so forth.
59

J. Comput. Civ. Eng. 1994.8:52-71.

 Zoning Data
Planners usually divide the project into different zones. The concept of
dividing a project into zones is usually used to concentrate closely on the
sequencing of activities in the zone as a separate subnetwork. The planner
defines interdependencies between the different subnetworks, so relations
between the activities of different subnetworks can be defined. The sub-
networks and the interdependency among them are usually combined to
Downloaded from ascelibrary.org by University of Exeter on 07/14/15. Copyright ASCE. For personal use only; all rights reserved.

come up with the final plan. In KNOW-PLAN, the concept of dividing the
project into different zones is incorporated. The system allows the user to
define the zoning structure for the project. A zone is defined by asserting
the boundaries of its volume space. This approach allows the search space
for a solution to be reduced. The result is a reduction in the time spent by
the system during the reasoning process.
The central database of KNOW-PLAN is implemented using a relational
data modeling approach. Each entity type and entity relationship is defined
by a relation table as a database file (DBF). The KNOW-PLAN's entity-
relationship model consists of three different entity types: activity, class,
and zone. The different entity types are related according to four types of
relationships. There is one simple relationship in the model: between activity
entities and class entities. The functionality of this relationship is many-to-
many. The other relationships in the model are complex (involuted) rela-
tionships with a functionality of many-to-many. The first is between activ-
ities. This relationship represents the connection data between activities of
different classes. The second is between classes. This relationship defines
the dominating class between each two classes. The third is also between
classes. This relationship defines the special cases under which the domi-
nating class relationship is not valid.

Other Knowledge
The KNOW-PLAN system can be expanded to allow additional knowl-
edge to be input and to be used during the reasoning process. The following
is a list of such potential knowledge:

Constructability Review Knowledge

By utilizing the path-finder, an AI-based system developed by Bechtel
(Morad et al. 1992), the user can check if a specific planning scenario is
feasible or not. This process is performed interactively by the user through
the simulation of physical object movement from a specific position to a
final position in the 3-D computer model using W A L K T H R U . The outcome
of the simulation process is a list of constraints to be considered during the
planning process.

User-Defined Knowledge
The user is allowed to define rules and constraints to be added to the
knowledge base in order to be considered during the reasoning process. The
user can add knowledge to the system in two forms. The first form is as
operational knowledge such as rules. The second form is as declarative
knowledge in the form of facts that defines a preferred precedence rela-
tionship among two activities. These facts are called sequencefacts in KNOW-
PLAN. These user-input facts combined with the sequence facts to be gen-
erated by the dynamic sequencer will be used in the refinement reasoning
process to come up with the final plan. The dynamic sequencing module
describes this issue in more detail next.
60

J. Comput. Civ. Eng. 1994.8:52-71.

 External Knowledge Sources
The KNOW-PLAN system is designed to accommodate any knowledge
source that follows the structure of the knowledge base of the system. The
only constraint on these knowledge sources is that they have to assert re-
lationships between activities in the form of sequence facts.
Fig. 1 depicts the data flow in and out of the central database in the
knowledge and data-entry module. The different kinds of knowledge passed
Downloaded from ascelibrary.org by University of Exeter on 07/14/15. Copyright ASCE. For personal use only; all rights reserved.

or entered into the dynamic sequencer include: system-defined geometric

knowledge, constructability knowledge, user-defined knowledge, and ex-
ternal knowledge sources. Fig. 2 includes the data and knowledge used by
the system to generate a project plan for the sample culvert project.

DYNAMIC SEQUENCING MODULE

The dynamic sequencing module is the centerpiece of the system. The

dynamic sequencer takes as input a description of the facility to be planned
as defined by the extracted geometric data from a 3-D computer model. It
progresses by using the extracted data in conjunction with geometric knowl-
edge and other knowledge to come up with a project plan. The plan gen-
eration process is based on a reasoning process that attempts to satisfy all
the constraints defined by the different knowledge sources in the system
based on the priority of these constraints.
The dynamic sequencer is basically a geometric-based planning system.
The sequencer adopts many AI concepts including: knowledge-based ap-
proach, least-commitment principle, state-space representation, and object-
oriented programming approach. The dynamic sequencer is implemented
using the ART-IM toolkit and runs under a DOS environment (ART-IM
manuals 1989).
The primary sources of reasoning knowledge in the dynamic sequencer
are the geometric data and the class knowledge stored in the central da-
tabase. The system incorporates external data interface rules to automati-
cally map the data and knowledge stored in the central database to the
schemata (frames) structure of its knowledge base. The dynamic sequencer
utilizes the two primary types of knowledge: declarative and operational
knowledge. The declarative knowledge represents the refined data that
describes the different classification of entities within the problem domain.
This kind of knowledge represents activities, classes, zones, and relation-
ships among them. During the reasoning process, the concluded relation-
ships among activities are asserted as declarative knowledge in the form of
sequence facts in the knowledge base. The operational knowledge consists
of rules that perform the actual reasoning process.
The dynamic sequencer's reasoning process is divided into two main phases
as shown in Fig. 3. The first phase involves the reasoning process to generate
relationships among the different activities of the project. These relation-
ships are based on the different constraints defined by the knowledge that
resides in the knowledge base. The second phase involves refining the gen-
erated relationships to optimize the project plan network. This reasoning
process is called the refinement reasoning process. Also, the second phase
performs schedule calculations based on the finalized project plan.

Phase 1
In the first phase, different kinds of rules are executed. There are rules
that perform external data interface. Other rules allocate activities into
61

J. Comput. Civ. Eng. 1994.8:52-71.

 Downloaded from ascelibrary.org by University of Exeter on 07/14/15. Copyright ASCE. For personal use only; all rights reserved.

External Data I n t e r f a c e Files

|
3ene
Sequ
D~,
User

P.quence facts

J. Comput. Civ. Eng. 1994.8:52-71.

~gicaJlyviewed
by user as
networks

Jrc~8

"IG. 3. Dynamic Sequencing Module

 zones. The main rules in this phase are the rules that are executed to perform
the geometric reasoning process. Finally, additional rules are executed to
conclude relationships between activities based on user-defined constraints,
constructability constraints, or other knowledge sources constraints.
In order to make the dynamic sequencer open to the external computing
environment, an external data interface is implemented in the dynamic
sequencer by a set of rules. Each rule provides access to an external ASCII
Downloaded from ascelibrary.org by University of Exeter on 07/14/15. Copyright ASCE. For personal use only; all rights reserved.

file. The ASCII files contain the different instances of the entity type or
entities' relationships that reside in the central database. Each rule performs
the necessary interface functions that create or modify the schemata struc-
ture of the concerned entity type or entities' relationships instances as they
exist in the knowledge base. Therefore, the activities, zones, classes, activity/
class relations, activity connections, dominating class relationships, and spe-
cial class relationships schemata are created during the external data inter-
face process.
In KNOW-PLAN, a physical component is considered to be located in
a specific zone if it partially or entirely overlaps with the zone's boundaries.
The dynamic sequencer uses the zone allocation rule to assert the zones in
which a component is located in or the zones with which it shares a surface.
KNOW-PLAN uses the geometric knowledge about the different compo-
nents of the designed facility and their associated classes during the reasoning
process to generate a possible project plan.
The geometric reasoning process is based on the conclusion of spatial
relationships of the project components in conjunction with the knowledge
that define the classes, direction of installation of components of the same
class, direction of installation of components of different classes, and con-
nection types between components of different classes. The geometric data
comparison between two components is performed by the geometric rea-
soning rules. These rules define which component among any two com-
ponents is higher in the x-direction, which is higher in the y-direction, and
which is higher in the z-direction. This conclusion is used in a geometric
reasoning rule to assert the precedence relationship between the two activ-
ities associated with the two components in the form of sequence facts.
There are three geometric reasoning rules that assert precedence rela-
tionships among the activities. The scope of each rule is as follows:

1. Rule 1: for relationships between activities of the same class in the

same zone.
2. Rule 2: for relationships between activities of the same class in different
zones.
3. Rule 3: for relationships between activities of different classes and
located either in the same zone or different zones.

In the case of components that are of the same class and their associated
components are located in the same zone, the rule starts by checking if the
two components are overlapping or sharing a common surface. If so, the
rule finds which component is higher in each direction of the three coor-
dinate axes: x, y, and z. Using the knowledge stored about the direction of
installation and the priority of installation of the activities' class, the rule
asserts a sequence fact that defines a possible precedence relationship be-
tween the two activities. In the case where two activities of the same class
and their associated components are located in different zones, the same
procedure used in the first rule is used. However, the rule checks only if
63

J. Comput. Civ. Eng. 1994.8:52-71.

 the two components share a common surface. If it finds a common surface,
the rule proceeds with the remaining actions. The right side of Fig. 4 illus-
trates an example for the geometric reasoning process to assert precedence
relationships between two activities, ACT[l] and ACT[2], of the same class
(C1).
The third rule asserts precedence relationships among activities of dif-
ferent classes whether the associated components are located in the same
Downloaded from ascelibrary.org by University of Exeter on 07/14/15. Copyright ASCE. For personal use only; all rights reserved.

zone or different zones. The precedence relationship between two activities

of different classes can be a function of the types of connection that exist
between their associated components and the two classes to which they are
related. For example, if the first activity is of class "mechanical" and the
second activity is of class "masonry wall," the mechanical component will
be installed before the wall. This is valid if the mechanical component is a
pipe and it is connected with the wall with an "embeddecl_in" type of
connection. However, the wall should be installed first if the mechanical
component is equipment to be "supported_by" the wall. The rule first checks
which component is higher in each direction: x, y, and z. Then it checks if
there is a special class relationship defined in the knowledge base that
matches the spatial relationship between the two components. If a special
class relationship exists, the direction of installation between the two classes,
to which the two components are related, is used to assert the precedence
relationship between the two activities. If no special class relationship exists,
then the dominating class relationship between the two classes is used to
assert the precedence relationship. The left side of Fig. 4 illustrates an
example for the geometric reasoning process to assert the precedence re-
lationship between two activities, ACT[2] and ACT[3], of two different
classes, CI and C2, respectively.
The geometric reasoning rules assert sequence facts in the knowledge
base to represent the concluded precedence relationships among the activ-
ities. The sequence facts have a relation name and a sequence of four A R T
objects. The relation name is "Execute." The first A R T object represents
the preceding activity. The second A R T object represents the succeeding
activity. The third A R T object represents the network to which the prec-
edence relationship belongs. All the facts that are generated by the geo-
metric reasoning rules are asserted as part of the geometric network that is
called " G E O M . " The last A R T object in the sequence represents the prior-
ity of the precedence relationship. The priority value of each precedence
relationship plays an important role in the process of combining the rela-
tionships of different networks into a final network at the last stage of the
dynamic sequencer's reasoning process.
The geometric reasoning process asserts sequence facts with a priority
value of 50. This priority value is expected to be the lowest value for the
sequence facts in the knowledge base. A typical sequence fact asserted by
the geometric reasoning rules may appear as follows:
(Execute ACT-1 ACT-2 GEOM 50)
In KNOW-PLAN, the user is allowed to define rules to define specific
constraints or requirements as separate knowledge sources. The outcome
of the constructability checking can be incorporated in the dynamic se-
quencer in the form of rules or directly asserted as sequence facts. The user
is also allowed to assert directly sequence facts to be considered during the
final reasoning process to generate the combined network. The only con-
64

J. Comput. Civ. Eng. 1994.8:52-71.

 Downloaded from ascelibrary.org by University of Exeter on 07/14/15. Copyright ASCE. For personal use only; all rights reserved.

|
~'lass
Class C1 & C2 Class RelatJo~ips Class Class "CI" Direction of Installation
| Class

Diri~dOh of lnstallation:in~:

Th~ Y Direction : :: . [ N_ [
: [ ThdZDir~ction:,:. I N [

Priority o f DirectiOn o f Installa"tion:

First Pn'otity ;:. : ix[
Second Priority

SpaU'al Relationships Cases Spatial Relationships Cases

J. Comput. Civ. Eng. 1994.8:52-71.

I' t 't 't ,t ,t
Installation ~-~uen~

t t t t t t

FIG. 4. Geometric Reasoning Based on Class Relationships

 straint on the rules that are defined by knowledge sources other than the
geometric reasoning rules is to assert sequence facts in the form of:
(Execute ACT~ ACTj Network-Name Priority)

The different sequence facts are grouped logically into networks that
define the source of the facts. The source of the facts can be from geometric
reasoning, constructability checking, user-defined knowledge, and so forth.
Downloaded from ascelibrary.org by University of Exeter on 07/14/15. Copyright ASCE. For personal use only; all rights reserved.

The combination of these networks represent the initial project plan.

Phase 2
The second phase of the dynamic sequencer takes as input all asserted
sequence facts. These facts are processed by the dynamic sequencer to come
up with the final project plan by refining the asserted plan. The purpose of
the refinement reasoning process is to eliminate redundancy, conflict, im-
plied logic, and loops. The dynamic sequencer has a set of rules that perform
this process. This set of rules can be viewed logically as an additional knowl-
edge source that knows how to refine a project plan.
The preliminary logic refinement is the first step in the refinement rea-
soning process. This is handled by two rules during the dynamic sequencing
process. The purpose of the first rule is to retract the sequence facts of low
priority that are implied by other high-priority sequence facts. The purpose
of the second rule is to retract all sequence facts of low priority that are
conflicting with other high-priority sequence facts.
At the current stage of the refinement reasoning process, after the pre-
liminary refinement logic, the knowledge base contains many sequence facts
that describe the potential precedence relationships among the different
activities of the project. The primary purpose of the refinement reasoning
process at this stage is to manipulate these facts to come up with a final
project plan that has no conflict.
The state-space transformation representation concept is explicitly uti-
lized in the refinement reasoning process. The refinement reasoning process
attempts to refine the defined precedence relationships that reside in the
knowledge base so conflicts are avoided. There are two types of conflicts
that could arise by asserting a precedence relationship in the final plan:
implied logic and loops. The process starts with the higher priority prece-
dence relationships and asserts them as permanent members of the final
plan network. The process proceeds by evaluating lower-priority precedence
relationships. The evaluation determines whether or not the selected rela-
tionship is to be asserted as a member of the final plan network. If a conflict
is expected to arise from asserting it in the network, the precedence rela-
tionship will not be asserted. If no conflict is expected, it will be asserted
(i.e., it is a member of the final plan network). The process proceeds by
selecting precedence relationships according to their priority values. This
process is performed at different cycles called priority cycles.
There is one primary action that performs the transformation process in
the state space, thus moving the search from one state to another. The
action is to select a precedence relationship from the potential precedence
relationships in the knowledge base, then to check if the selected precedence
relationship provides any conflict with other precedence relationships that
are already asserted as members of the final plan network. Based on the
outcome of the checking process, changes in the knowledge base occur.
These changes define a new state description in the state space.
There are many rules in the dynamic sequencer that are concerned with
66

J. Comput. Civ. Eng. 1994.8:52-71.

 this stage of the reasoning process. Some of these rules perform the actual
final refinement reasoning process. Others are included to lend support and
pass control from one rule to another. The primary outcome of the reasoning
process is a final project plan, a network of activities with precedence re-
lationships among them.
The next step is to calculate the schedule dates: early start, early finish,
late start, and late finish. An object-oriented programming approach is
Downloaded from ascelibrary.org by University of Exeter on 07/14/15. Copyright ASCE. For personal use only; all rights reserved.

adopted to perform the basic schedule calculation task. The dynamic se-
quencer sends messages to the activities from an action side of rules defined
to perform the forward and backward schedule calculation task.
The last step to be performed by the dynamic sequencer is creating two

il ES ILS IE F ILF I ~
No~:
A user-defmed sequence fact has been incorporated in generating this network
(Execute ACT-3 ACT-2 USER 60)
FIG. 5. Generated Plan of Sample Culvert Project

67

J. Comput. Civ. Eng. 1994.8:52-71.

 Downloaded from ascelibrary.org by University of Exeter on 07/14/15. Copyright ASCE. For personal use only; all rights reserved.

J. Comput. Civ. Eng. 1994.8:52-71.

)lay-back
~ Routine
Modifications
I
New Facts and Rules
Additional Knowledge Source)

FIG. 6. Visual Simulation Module

 Downloaded from ascelibrary.org by University of Exeter on 07/14/15. Copyright ASCE. For personal use only; all rights reserved.

3D Computer Modoling Module Knowledge and Data Entry Modulo

Activity Data
Extracted Class Data
;D Computer 3eometric Data Dominating Class Relations
Modeling Special Class Relations
Activity Class Relations Jser
t.ctivity Connection Relations
Zoning Data

External Data

J. Comput. Civ. Eng. 1994.8:52-71.

Lo~c ]~des Knowledge
VisualSimu/ation Module Dynamic Sequencing Module
Sources
FIG. 7. )verall Picture of Know-Plan Modules
 ASCII files that contain the precedence relationships data among different
activities and the schedule dates of different activities that represent the
final project plan. Fig. 5 illustrates the activity-on-node network that rep-
resents the generated plan by the dynamic sequencer for the sample project.

VISUAL SIMULATION MODULE

This module performs the visual simulation of the construction process
Downloaded from ascelibrary.org by University of Exeter on 07/14/15. Copyright ASCE. For personal use only; all rights reserved.

based on the plan generated by the dynamic sequencer. This module pro-
vides a practical approach for representing project plans in a simple and
realistic fashion. The strength of the KNOW-PLAN system comes from its
ability to integrate KBES technology with 3-D computer modeling to gen-
erate and visually simulate project plans.
The visual simulation module takes as input the two ASCII files generated
by the dynamic sequencer. The primary component of the visual simulation
module is the postprocessor of the central database. The postprocessor is
a processing environment that creates the necessary data needed to perform
the visual simulation process. The postprocessor reads the two ASCII files,
then creates new database files and modifies existing database files based
on the data included in the ASCII files. Finally, it creates a record file.
The record file is the basic element of the interface between the central
database and W A L K T H R U . It includes parameters that define the config-
uration of the display at each time frame. Time frames are called key frames
in the W A L K T H R U system. At each key frame of the record file, specific
viewing parameters are defined. These parameters define the x, y, and z
position of the body that represents the position of the user in the 3-D
computer model. The direction of travel, the head orientation, clipping
planes, and the perspective angle of the view at each key frame are also
included in the key frame parameters. Each key frame starts with an at-
tribute that defines the delta time since the last key frame.
The central database provides a data-entry form to be filled by the user
to define the viewing parameters for a specific record file. These parameters
are stored in a database file. The user is required to provide only one set
of viewing parameters (usually taken from an existing record file). This set
is used by a)l key frames in the record file.
The KNOW-PLAN system utilizes the record and replay functions of
W A L K T H R U to perform the visual simulation of the construction process.
The postprocessor uses the activities' schedule data to create a key frame
in the record file whenever an activity starts or an activity finishes. The
result is a record file that contains a sequence of key frames that represents
the time at which each activity is to be displayed during the visual simulation
process.
The visual simulation of the construction process provides a simulation
tool to present the sequence of installing the physical components of the
project on a graphics display. The simulation process shows when each
component should be installed based on the early or late schedule of the
project. The simulation process is viewed according to a time scale that is
controlled by the user. Fig. 6 illustrates the data flow in this module. Fig.
7 provides an overall picture of the different modules of the system.

CONCLUSION
CAD and 3-D computer modeling technologies provide powerful graphics
support for construction planning. Knowledge-based expert systems provide
capabilities to automate the process of generating project plans.
70

J. Comput. Civ. Eng. 1994.8:52-71.

 KNOW-PLAN is an advanced planning and scheduling system that in-
tegrates 3-D computer modeling technology with AI technology and over-
comes some of the limitations that exist in current planning techniques. The
system generates construction plans using a geometric-based reasoning ap-
proach. The generated sequence is based in part on the geometric data that
define the spatial relationships among the various components of the de-
signed facility that is acquired from a 3-D computer model of the facility.
Downloaded from ascelibrary.org by University of Exeter on 07/14/15. Copyright ASCE. For personal use only; all rights reserved.

The generated sequence also depends on the different knowledge rules

defined in the knowledge base. The planning process is enhanced by pro-
viding visual tools to simulate graphically the construction process.
Project planning can greatly benefit from the integration of 3-D computer
modeling and knowledge-based systems to automatically generate project
plans and to visually simulate the construction process. More accurate con-
struction schedules will be generated. Greater understanding and awareness
of the project by management can be achieved by visually simulating the
construction process. During actual construction, the simulation process will
help site personnel to visualize what should be constructed. Visual simu-
lation of the construction process will be a valuable tool to support and
defend construction claims in the future.
In summary, the KNOW-PLAN system provides an additional step in
the direction of automating the process of generating project plans. This
step is based on the integration of advanced technologies to automatically
generate project plans based on the geometric data of the different com-
ponents that comprise the project and their associated classes.

APPENDIX. REFERENCES
ART-IM manuals. (1989). Inference Corp., Los Angeles, Calif.
Clough, R., and Sears, G. (1979). Construction project management. John Wiley and
Sons, New York, N.Y.
Dym, C., and Levitt, R. (1991), Knowledge-based systems in engineering. McGraw-
Hill, Inc., New York, N.Y.
Encarnacao, J., Schuster, R., and Voge, E. (1986). Product data interface in CAD/
CAM applications. Springer-Verlag, Berlin, Germany.
Morad, A., and Beliveau, Y. (1991). "Knowledge-based planning system." J. Constr.
Engrg. and Mgmt., ASCE, 117(1), 1-12.
Morad, A., Cleveland, A., Beliveau, Y., Francisco, V., and Dixit, S. (1992). "Path-
finger: AI-based path planning system." J. Comp. in Cir. Engrg., ASCE, 6(2),
114-128.
Skolnick, J., Morad, A., and Beliveau, Y. (1990). "'Development of a CAD-based
construction visual schedule simulation system." Proc. PMI '90 Conf., Calgary,
Canada, 334-340.
"WALKTHRU Bechtel's 3D simulation system user's manual." (1991). Bechtel
Power Co., Gaithersburg, Md.

71

J. Comput. Civ. Eng. 1994.8:52-71.