Engng Applic. Artif. lntell. Vol. 8, No. 4, pp.

435-447, 1995
Elsevier Science Ltd. Printed in Great Britain
Pergamon
0952-1976(95)0~19--4

Contributed Paper

A Process Model for Knowledge-based

Concrete Bridge Design
H. SHIVA KUMAR
Indian Institute of Technology, India

C. S. KRISHNAMOORTHY
Indian Institute of Technology, India

N. R A J A G O P A L A N
Indian Institute of Technology, India

(Received November 1994; in revised form February 1995)

Modeling the design process is an important step in knowledge-based system (KBS) development.
This involves a thorough examination of the application domain, and an identification of the required
artificial intelligence techniques for the development of KBS. This paper presents one such research
programme, carried out in the domain of concrete road bridge design. KBSs for bridge design have
been traditionally implemented as production rule systems. This paper argues that an integration of
Al-based problem-solving techniques is necessary to address the various tasks of bridge design. It has
been demonstrated that a design process model based on rule-based inference, synthesis and critical
evaluation techniques is required to address the knowledge-intensive tasks of site selection, bridge
layout planning, conceptual design and preliminary design of concrete road bridges, in an efficient
manner.

Keywords: Knowledge-based system, process model, rule-based inference, synthesis, criticism,

bridge design.

I. INTRODUCTION I.I. Prior work

Since the advent of knowledge-based system (KBS) Various process models have been proposed for
techniques for solving complex problems, various diffi- solving different design problems. In these, the design
cult tasks of engineering design domain have success- tasks are formulated as search problems, and some
fully been computerized, l' 2 In a given application do- control strategies are applied to search through the
main, the tasks that require expertise for problem- state space and arrive at one or more solutions. The
solving are identified, and the knowledge required is most basic control strategy is generate-and-test, where
acquired from acknowledged experts. The acquired all possible states are generated in a systematic manner
knowledge is suitably formalized and a KBS is built. and then each one is tested to see if it is a goal state.1
The main step in the development of a KBS is modeling Later, the generate-and-test model was improved and
the design process. This involves a careful examination modified as hierarchical-generate-and-test, in which par-
of the application domain and determining the tial solutions are tested to avoid searching those solu-
appropriateness/applicability of various available artifi- tion paths that do not yield solutions. This model
cial intelligence (AI) techniques. The acquired know- implicitly assumes that partial solutions can be gener-
ledge is then organized on the basis of the design ated and tested, and that pruning can be done
process model, and represented suitably for the deve- correctly. 3 The design process model of Bedard and
lopment of the KBS. Ravi 4 is one such example, in which all possible solu-
tions are generated at some relatively high level of
abstraction. Inferior solutions are heuristically elimi-
Correspondence should be sent to: Professor C. S. Krishnamoorthy,
Department of Civil Engineering, Indian Institute of Technology, nated, and the rest are taken up for building the
Madras 600 036, India. solution further at the next level of abstraction. The
435
436 H. SHIVA KUMARet al.: KNOWLEDGE-BASEDCONCRETE BRIDGE DESIGN

process of elimination and generation is continued till the domain of bridge design. First, a few related efforts
the final solution is obtained. carried out in the area of knowledge-based systems for
EDESYN 5 is a domain-independent implementation bridge engineering are presented. Various subtasks
of the hierarchical generate-and-test model, and has that constitute the bridge design and the problem-
been used to develop design expert systems like solving methodologies adopted in the respective sub-
STRYPES, STANLAY and FOOTER that form part tasks are then discussed. Based on these methodolo-
of a larger design environment for integrated building gies, the required artificial intelligence (AI) problem-
design. 6 Sause and PoweU7'8 have presented an organi- solving techniques are identified. A process model for
zational model of the structural engineering design the design of concrete road bridges is then presented.
process called the multi-level, selection-development The salient features of the proposed process model are
(MSD) model. The model organizes the design process illustrated with the aid of an application example.
by identifying generic subproblems to be solved at Finally, the conclusions derived on the basis of this
different levels of abstraction, and provides a mecha- experience are enumerated.
nism for transfering information between subproblems.
A variation of the generate-and-test that has been suc-
2. RELATED EFFORTS IN BRIDGE
cessfully applied to design problem solving is the con-
ENGINEERING DOMAIN
straint satisfaction model. 1 In this approach, the evolv-
ing design is tested to see if constraints are violated. BDES ~3is a production-rule-based system limited to
The focus here is on how the constraints are formulated the choice of bridge deck for American practice using a
and propagated as the design evolves. Luth et al. 9 have given geometry and pier location. Another expert
proposed a model for automating conceptual structural system developed by Burgoyne and Sham TM for the
design, where reasoning is done through formulation, choice of prestressed concrete bridge decks is modeled
propagation and satisfaction of constraints. The model using Prolog-based production rules. Adeli and
has been used to implement a floor framing generator Balasubramanyam 15-18have proposed a coupled (com-
for steel office buildings that are rectangular in plan. 10 bination of rules and procedures) model for the optimal
Blackboard models are widely recognized as a means design of bridge trusses. Their system, BTEXPERT
for addressing problems that require multidisciplinary has machine learning capabilities in-built into it, and
efforts to solve them. GUIDE, H for multidisciplinary enhances the knowledge-base when a novel situation is
probem solving, adopts this approach in its effort to encountered. RCABE, ~9an expert system for the pre-
establish proper communication between the partici- liminary design of bridges, is built using a production-
pating agents during the solution process. The inte- rule-based model and is limited to small-to-medium
grated building design environment (IBDE 4) project is span bridges and major box and pipe culverts, for
similarly concerned with the cooperative integration of which standarized components are available. Moore
the architecture, engineering and construction descip- and Miles 2 have addressed the conceptual design
lines to automate the entire building design process. (including the bridge layout planning) of bridges,
Ahmed et al. 12have described a model for cooperative within a limited scope of roads over bridges (typically
problem solving on distributed workstations. In this motorways) having four spans, with the aid of a
model the blackboard is implemented as an object- Prolog-based production-rule model. They have
oriented database, through which all participating improved the efficiency of the system by adopting
knowledge sources communicate. meta-rules to prune the search space (by controlling the
backtracking). SUPEREX, 2~ an expert system for
1.2. Motivation selecting the type of superstructure, is the latest pub-
Although various design process models have been lished KBS which is built using a coupled model as in
proposed for various application domains, choosing an the case of BTEXPERT. It uses a production-rule-
appropriate model or designing a suitable process based grading for evaluating and selecting generated
model for any particular application domain requires a solutions. A number, representing the quality of the
detailed study of the domain under consideration. solution according to some criterion, is assigned to each
Various subtasks and their methodologies are identi- of the generated solutions and (based on the user-given
fied during this study. Based on the identified methodo- preferences for each of the evaluation criteria) the
logies and the nature of the information flow required overall grading of the solution is calculated for final
from one subtask to another, a design process model is selection.
arrived at. A similar modeling strategy has been All the above systems are modeled using production-
adopted by Sause and Powell to come up with the rule-based techniques. The development efforts
MSD 7.s model for building frame design. It is very clear involved in production-rule-based systems are tremen-
from this example that the design process differs from dous. Hence, in the present work, an attempt is made
one application domain to another, and depends on the to identify suitable AI problem-solving techniques,
design practice adopted at that particular place. based on the methodolgies adopted in practice. A
This paper presents one such exercise, carried out in design process model is proposed accordingly, to build
 H. SHIVA KUMARet al. : KNOWLEDGE-BASEDCONCRETEBRIDGE DESIGN 437

a KBS that enables the knowledge engineer to organize a KBS. However, in the paper, the discussions are
the knowledge in an efficient manner, such that the limited to the selection based on the critical evaluation
maintenance of the system becomes easy and reduces of chosen sites for serviceability and economy.
the efforts involved in implementation. These aspects The factors affecting the selection of a site are
are discussed in the following sections. grouped under the following four major categories in
Indian conditions.
3. BRIDGE DESIGN PROCESS
(1) Alignments: the bridge alignment with respect
Bridge engineering is an important activity in civil to the obstacle and the approach alignment
engineering, and essentially involves the construction with respect to the existing road are con-
of a passage for transportation across an obstacle. It sidered here.
includes the tasks of (1) selecting (a) suitable bridge (2) Geological conditions: the type of foundation
construction site(s) from the alternative sites chosen, strata available, depth of the foundation strata
(2) planning the layout of the bridge at the selected and the existence of fault formations in the
site(s), (3) choosing suitable structural forms for the strata are grouped into this category.
bridge layouts, (4) preliminary dimensioning of the (3) Construction facilities: availability of labour,
chosen structural forms, (5) analysis and design of the construction materials and construction equip-
structures, (6) constructing the bridge and (7) maintain- ment fall into this category.
ing the structure over its life time. The magnitude of (4) River bridge aspects: aspects such as tributary
expertise required for creating a system which would confluence and underwater work required are
cope with any given situation is vast. Moreover, this considered in this category.
information is very poorly documented, and extracting
such knowledge from experts is extremely difficult.2 The above aspects of the four categories are further
Patridge22 has pointed out that trying to develop a classified as shown in Table 1 for site evaluation. This
system which caters for all the requirements of the classification reflects the quality of each of these aspects
practical design is too ambitious, and such attempts that will ultimately be evaluated.
would result in unusable systems. Since the usability of
the system is of paramount importance, in the present Table 1. Classificationof importantaspectsof site selection
work the scope is limited to the first four tasks only.
The problem-solving methodologies required for the Category Aspect Classification
various subtasks of these four tasks, that require KBS Alignments Bridge Alignment Square
applications, are given in the following sections. Moderate
High
3.1. Site selection Severe
Approach Alignment Straight
The objective of this task is to collect the relevant Medium
data about all the possible sites for bridge construction, Curved
and then to choose the appropriate one on the basis of Sharp
Geologicalconditions Type of strata Hard rock
an evaluation of the alternative sites. This is an import- Weathered rock
ant task in the overall bridge design, since the economy Sandysoils
of the bridge construction depends heavily on the site Laterites
Oay
chosen. There are a number of factors that have to be Fault formation No faultzone
taken into account, such as the characteristics of the Safe zone
sites, the economic benefits of constructing the pro- Unsafe zone
Depth of strata Best
posed bridge at the various locations, existing align- Good
ments of roadways and highways, etc. Average
The site-selection process starts by first choosing a Bad
Constuctionfacilities Labour at the site Very near
few bridge locations on the topographical map of the Near
area and then by visiting each of these locations to Away
collect the data for selection. Inappropriate sites are Far away
Material at the site Very near
eliminated on the basis of the field and laboratory tests. Near
For the chosen sites, various other aspects governing Away
the serviceability and economy of bridge construction Far away
Equipment at the site Available
are processed and classified into various categories for Not available
critical evaluation and final selection of appropriate River bridge aspects Tributary confluence None
sites. After the critical evaluation of the various sites, Away
Close
final selection of a site is made for bridge construction. Very close
Of these, the task of choosing sites on the basis of field Under water work Least
and laboratory test results, and selection based on the Moderate
Excessive
critical evaluation, require expertise and hence require
438 H. SHIVA KUMARet al. : KNOWLEDGE-BASEDCONCRETEBRIDGE DESIGN

3.2. Bridge layout planning of thumb, empirical formulae and codal provisions.
The objective of this task is to arrange the positions The heuristics help in arriving at the preliminary values
of abutments and piers such that the resulting number of the dimensions of the structural components without
of spans and span lengths are satisfactory. Factors such the use of the rigorous iterative computational pro-
as site topology, constructability, economy and struc- cedures of structural mechanics that would otherwise
tural safety play a major role in this task. The abutment be required.
locations are chosen on the basis of the amount of Based on the above preliminary dimensions, the
cutting/filling costs this entails. After deciding the abut- weight of the structure can be estimated. The prelimi-
ment locations, the pier locations are based on the well- nary weights thus determined will give an idea about
known bridge economy criteria, viz. cost o f super- the cost of the structure, and on this basis it is possible
structure is equal to the cost o f substructure. On this to eliminate uneconomical solutions. Hence, a compar-
basis, a heuristic is formed for deciding the pier ison of the various alternative preliminary designs is
locations. That is, the piers are placed such that the made at this stage, and those preliminary designs that
space between two piers or pier and abutments is 1.5 to are likely to be economical can be retained for analysis
2.0 times the total height of the structure. Once the pier and detailed design.
locations (and hence the span lengths) are thus
decided, the following checks are made to test the 4. IDENTIFICATION OF PROBLEM-SOLVING
validity of the bridge layout: TECHNIQUES
Whether the clearance requirements under- It is very clear from the above section that the bridge
neath the bridge for traffic flow, if any, are design tasks presented there cannot be efficiently
satisfied. modeled using production rules alone. Specialized
Whether the waterway requirements, in the techniques need to be adopted in conjunction with
case of river bridges, are satisfactory in order to production rules for addressing the issues in a compre-
keep the velocity, scour and afflux under hensive manner. The AI-based techniques for various
control. tasks of bridge design are proposed in the following
Whether the existing foundation strata are cap- sections.
able of taking the loads from the piers.
4.1. Site selection
When any one of the above is not satisfied, the bridge
layout is modified immediately to suit the particular As described before, this process involves classifying
requirement in such a way that the other constraints are various characteristics of the site, and then selecting
not violated. those sites that meet the requirements to a greater
degree. The classification task can be modeled using
3.3. Conceptual design rule-based techniques, 23'24 where the required classifi-
In the conceptual design phase, a structural designer cation criteria can be efficiently represented in the form
arrives at a suitable structural system that meets the of production rules.
requirements. For the chosen bridge layout, there can Various sites would meet the economic and service-
be a number of different possible structural systems. ability requirements with varying degrees of satisfac-
Therefore, bridge design experts choose structural tion. The ideal way to make a selection would be to
systems by considering the feasibility requirements. evaluate each of the competing sites on the basis of the
These feasible structural systems are then evaluated individual selection criteria for each of aspects listed in
from aesthetic, economic and maintenance viewpoints. Table 1, and assign a rating that reflects the site's
The symmetry of the selected structural systems is degree of acceptability from the particular viewpoint.
evaluated from the aesthetic viewpoint. Construction All such individual ratings can be combined later to
costs such as the cost of formwork, centering, launch- arrive at an overall rating of the site. The one with the
ing, etc., are evaluated from the economic viewpoint. highest overall rating then becomes the most obvious
Finally, the maintenance aspects of the bridge are choice.
evaluated, on the basis of the type of structure chosen, The overall rating of a site depends on the relative
the construction adopted and number of joints and importance of each aspect to be assessed. That is, each
bearing. aspect considered has a different bearing on the overall
rating of the sites. Note that the classified aspects listed
3.4. Preliminary design in Table 1 can be viewed as a hierarchical network, as
The objective of this task is the preliminary sizing of shown in Fig. 1. Such arrangements can be represented
structural components. The depth of the deck slab, the efficiently using a frame-based representation scheme.
positions and dimensions of the main and the cross The four major categories are represented as compo-
girders, the dimensions and the cross-sectional areas of nents (frames) of the main system site, and the aspects
other load-transferring systems are fixed at this stage. to be evaluated as their attributes. Since the major
This stage involves considerable use of heuristics, rules categories can also have a different bearing on the
 H. SHIVA KUMARet al.: KNOWLEDGE-BASEDCONCRETEBRIDGE DESIGN 439

site

ts

bridge approach / ~ tributary underwater

alignment alignment/ ~onfluence work

geological conditions construction facilities

type of depth of fault labour material equipment

strata strata formation at site at site at site
Fig. 1. Representation of site data.

overall rating of the solution, in a hierarchical frame- layout planner, traffic engineer, hydraulics expert and
based representation, the components are also asso- foundation expert. The layout planner proposes a
ciated with different relative importance factors. bridge layout based on the economic criteria. This
Therefore, while determining the overall worth of the layout will be validated immediately by the traffic
site for bridge construction by combining the individual engineer who checks whether the layout satisfies the
ratings, different weightings are given to each of the clearance requirements. If the layout does not satisfy
components and the aspects to be evaluated. For exam- the requirements, the traffic engineer suggests the
ple, in Indian conditions, the category alignment is modifications to be made to the proposed layout. In
twice as important as any one of the other three case of river bridges, the hydraulics expert tests the
categories. Therefore, a 40% weight is given to the proposed layout for waterway requirements and sug-
alignments, and 20% each to the remaining three gests modifications if the proposed bridge layout does
categories in the determination of the overall worth of not satisfy waterway requirements. Similarly, in addi-
the site. Similarly, within the category, construction tion to the above two, the foundation expert checks
facilities, the labour aspects are considered twice as whether the foundation strata are capable of supporting
important as the other two. A weighting of 50% is the bridge, or if suitable foundations can be adopted to
given to the labour aspects, and 25% each to the other transfer the load safely, and suggests modifications
two. accordingly. Thus satisfying all such requirements, a
From the above critical evaluation process, it is clear mutually agreeable bridge layout will be decided. The
that a methodology consisting of rating individual layout of a bridge with n spans is shown in Fig. 2.
aspects and determining the overall worth of the solu- The bridge layout planning is similar to the multi-
tion on the basis of their relative importance is found to disciplinary problem-solving discussed above. A solu-
be appropriate. tion is proposed and immediately evaluated to check its
Therefore, the two AI-based problem-solving tech- validity from various viewpoints. Hence, AI techniques
niques that are ideal for the site selection task are: that emulate the multidisciplinary problem-solving are
appropriate for bridge layout planning. The knowledge
(1) rule-based inference for classifying the site
required for layout planning can be represented in the
characteristics, and
form of production rules, and the clearance, hydraulic
(2) critical evaluation of hierarchically represented
and foundation requirements as constraints. The solu-
objects which involves the determination of the
tion can be represented as a dependency net, consisting
overall worth of solutions on the basis of the
of various interconnected bridge layout planning steps,
ratings of individually assessed aspects and
for enabling the modifications and their global reper-
their relative importance for selecting suitable
cussions. Therefore, the following AI-based techniques
bridge construction sites.
are required for emulating the multidisciplinary
problem-solving methodology.
4.2. Bridge layout planning
The process of bridge layout planning can be viewed (1) rule-based inference for layout planning,
as an interaction between four design experts, viz. (2) constraint posing for ensuring the suitability of
440 H. S H I V A K U M A R et al.: K N O W L E D G E - B A S E D C O N C R E T E B R I D G E D E S I G N

abutment[l] spnnlZl spnn[n-ll abutmentlll

spanlll spanlnl
(SLID (Stll)

pier[l] '~ q pier[n-I]

I I (F[I]) (vt,-~])

abutment foundation[ll pier foundation[l] pier foundation[n-l] abutmentfoundation[2]

(AF[I]) (PF[I]) (PF[n-I]) (AF[2])
Fig. 2. Layout of a bridge with n spans.

the proposed bridge layout, and feasible combinations in a huge search space) is very
(3) dependency-based backtracking for maintain- difficult, and the maintenance (addition of new compo-
ing the compatibility between the modified nents, modifications, etc.) becomes extremely difficult.
parts and the rest of the solution. Since the number of feasible combinations depends on
the given context, formalizing the rule-based synthesis
4.3. Conceptual design becomes increasingly difficult.
As shown in the previous section, an expert gener- The degree of acceptability of a synthesized solution
ates various feasible conceptual designs and screens can be gauged by subjecting it to a series of tests that
them by critical evaluation to choose the appropriate examine its adherence to global objectives not con-
ones for preliminary design. The bridge can be viewed sidered at the solution-generation stage, such as econ-
as a hierarchical network of its component/ omics, aesthetics and maintainability. The critical
subcomponent systems. The hierarchical represen- evaluation of the synthesized conceptual design solu-
tation of the bridge shown in Fig. 2 is shown in Fig. 3. tions is similar to that of alternate sites, discussed
This type of organization of the system is termed a above. Various aspects of the conceptual design are
"decomposition". Each component has various attri- assessed individually and in various combinations, and
butes constituting it, and various possible values are appropriate ratings are assigned. These ratings are
associated with them. Therefore, arriving at a solution combined later to arrive at the overall rating of the
involves assigning a suitable value to each attribute of solution which reflects the degree of acceptabilty of the
the component systems. Incompatible assembly of synthesized solution. The component systems of the
component elements can be represented as elimination bridge shown in Fig. 3 have different weights associated
constraints. By putting together the component with them for determining the overall rating. Similarly,
systems, as described in the solution decomposition in the above evaluation, the total worth of the concep-
tree, all feasible solutions can be generated using a tual designs is determined with different relative
constraint-directed search. This process is termed importance factors for the aesthetics, economic and
decomposition-based synthesis. 5 As stated earlier, maintenance aspects to be rated for each component
there is no unique solution to the conceptual design system. According to the experts consulted, in Indian
problem. The decomposition-based synthesis approach conditions, a weighting of 80% is given for aspects
thus enables the generation of a large number of belonging to economy, and 10% each for the aspects of
potential candidates. the remaining two criteria.
Figure 4 shows a typical decomposition tree of a Therefore, the problem-solving techniques required
bridge structure consisting of the component/ for conceptual design are:
subcomponent systems, their attributes and the
(1) decomposition and constraint-based synthesis
respective possible values.
for feasible solution generation; and
Rule-based synthesis can also be adopted for the
(2) critical evaluation of hierarchically represented
conceptual design task. However, formalizing the
objects which involves the determinaton of the
knowledge in the form of rules (by figuring out all
overall worth of solutions based on the ratings
Bridge
of individually assessed aspects and their rela-
tive importance for eliminating inferior feas-
ible solutions.

4.4. Preliminary design

super structure sub structure foundation
As has been shown, the preliminary design task is
characterized by the use of rules of thumb gained
through years of experience. The rules of thumb
SillS[21...S[nl A[II1'[11P[21... P[n-l] A[21 ~[ll PF[ll PF[2]... PF[n-IIAF[21 provide a short cut to the cumbersome iterative process
Fig. 3. Representation of a bridge. of arriving at the dimensions of structural elements by
 H. SHIVA KUMAR et al. : KNOWLEDGE-BASED CONCRETE BRIDGE DESIGN 441

I simple
support ~ continuous
fixed
cast in_sltu
construction_type~ precast
slab
[Link] ~ t_beam
box_girder
solid
~ voided
spas cross section with bottom bulb
- without bottombulb
straigbt webs
inclined_webs
plain
reinforcement~ prestressed
bridg~

cast in_situ
~ ~
construction_type precast
tbstructure abutment ~~ _m_ . a. - s~ o a l rc 7
rc
pier ~ m a s o a r y

cast ln_situ
~ construction_type~ precast
foundations abutmentfoundation/shallow
.

~-deep

pier foundation ~ d e e p
shallow
Fig. 4. Decomposition tree for the conceptual design of concrete road bridges.

well-defined principles of structural mechanics. Various techniques that are required at this stage are:
preliminary designs thus arrived at are compared and
(1) rule-based inference for the preliminary di-
analysed according to least weight criteria to choose the
mensioning, and
one with the minimum material cost. This assessment
(2) comparative assessment for evaluating the pre-
involves assigning ratings to the preliminary designs on
liminary designs according to the minimum
a comparative basis, and differs from the two assess-
weight criteria.
ment techniques discussed above. Here, the one with
the minimum weight is given the maximum rating (a
value of 100) and the rest are given ratings on a 5. PROCESS M O D E L
comparative basis. Therefore, the problem-solving
From the above sections it is very clear that the
solution-generation process in bridge design follows a
sequential task-wise path in spite of its complex and ill-
structured nature. The proposed design process model,
synthesized 1 evaluated
solutions solutions viz. rule-based inference, synthesis and criticism (RSC)
model, for the concrete road bridge design is shown in
Fig. 5. In the RSC model, the expert knowledge is in
four forms, viz. (1) heuristic knowledge, (2) synthesis
knowledge, (3) evaluation (assessment) knowledge,
and (4) procedural (algorithmic/computational) know-
ledge. Heuristic knowledge is represented as rules, and
is used for decision making and overall process control,
. T rule-based synthesis and spot assessments. The synthe-
sis knowledge is in the form of a decomposition tree
Fig. 5. RSC model. and constraints. Evaluation knowledge is in the form of
442 H. SHIVA KUMARet al. : KNOWLEDGE-BASEDCONCRETE BRIDGE DESIGN

decomposition tree
and constraints heuristic knowledge
II evaluation criteria

I I I
FRAME-BASE
intermediate and ntl solutions

Fig. 6. Blockdiagramof KBS.

testing criteria and is used to evaluate the competing ment tools, DEKBASE 25 (Development Environment
alternative solutions qualitatively and quantitatively for for Knowledge-BAsed Systems in Engineering) has
screening. Finally, the fourth type of knowledge, viz. been chosen for this prototype KBS development.
procedural (algorithmic/computational) knowledge,
consists of knowledge pertaining to analysis, detailed
design, and other numerical computations. These are 7. APPLICATION EXAMPLE
represented as procedures in any standard procedural The application example presented here is a com-
programming language. piled form of a working session of the developed KBS
As the synthesis and critical evaluation are done by for the design of concrete road bridges. The example
different process engines (synthesis by rule-based infer- depicts the salient features of the RSC model presented
ence engine and synthesizer, and evaluation by a rule- in the previous sections. The application example is
based inference engine and critic), communication discussed in a task-wise manner, as in the previous
between these is very important. Besides, as the solu- sections while discussing the domain and identifying the
tion generation is carried out incrementally at various required A1 techniques.
design stages (such as site selection, bridge design The bridge construction location is shown in Fig. 7.
layout, conceptual design and preliminary design), As can be noticed, it is a road over a bridge for crossing
proper communication between these tasks is required. the proposed railway lines. There are eight broad-
In the proposed model, the communication between gauge tracks running on either side of the control cabin.
the process engines and tasks is maintained through The three possible bridge alignments are shown on the
framebases and global data. Each valid partial solution plan of the location. The first bridge alignment has an
is stored in a framebase and global data, and is access- angle of skew of 10, and the second 20. The third one
ible to the process engines and design tasks. The block is a square crossing. The approaches of the second
diagram of the KBS for concrete road bridge design alignment are straight, and those of the first and third
shown in Fig. 6 depicts the communication aspects of are curved. The first one is chosen such that the length
the RSC model. is less than that of the second, and its approaches are
less curved than those of the third.
The site selection, conceptual design and preliminary
6. K B S D E V E L O P M E N T T O O L design tasks carried out by the developed KBS are
From the above two sections, it is clear that the described in the following sections.
following facilities are required for KBS development.
7.1. Site selection
rule-based inference with backtracking facilities At this stage, the suitability of the three chosen sites
synthesis is determined by classification and evaluation as des-
evaluation cribed earlier. The information about the three possible
framebase-management system for proper com- alignments chosen for an over-bridge construction are
munication between various solution-building classified by the site information classifier as given in
stages and process engines Table 2. The various aspects given in Table 1 are
database-management system for storing static critically evaluated and the individual ratings are com-
design data bined to arrive at the overall ratings. The process of
procedural programming interface for computa- combining the various individual ratings of the aspects
tionaUy intensive tasks. being evaluated is explained in an earlier publication. 26
The results of the critical evaluation of the three sites
After a careful study of the available KBS develop- by the site evaluator are also given in Table 2. Based on
 H. SHIVA KUMAR et al. : KNOWLEDGE-BASED CONCRETE BRIDGE DESIGN 443

existing road ~ . !i i:

:iii!:
~iiii
proposed railway lines

iilllllliiiiiiiiiiiiii
of width 1.6Sm each @ 6.65 m e/c ~A

iiiiiii i/ i
I I I A . . . . . . . . .

.I I. I . I . I .I I. I . I . I .I I. I . I . I .I I. I . I . I . . . .
. . . . . . . . .

I I I i i i.i i i i i i i i i i i i i I I i i i i i i i i i
, . . . . . . . . . . . , . . . . , , . , . , , , , .
.. . . .. : / i i i i, i, i , i , i ,i l, l, l , l , l ,i ,i i, l . . . . . .

ii!ii!!!ii!!!!!!iiilliiiiiii! i ~ " ~ i i i i i i I I i i i j j j i i i i j j
I I I I I I I I I I I I I I I I I I I I I I I I I I I I I u~.'l i ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! !
[Link]
control cabin
i ( s ~ [Link] X 4.0m)
. . . . . . . . . . . . . . . . . . . . . . . . . . . . :)~
I I I I l i I I I I I I I I I I ] ] I I I ] I I I I I I:/llf ! ! . . ! ! ! ! ! ! [ ! [ [ ! ! [ ! ! ! ! ! ! [
I. .I .I .I . I . I . I . I . I . I . I . I . I. I. I. I. .I .I .I . I . I . I . I . I . I I I:~l~'i , I l l / l l l l l l l l l l l l l l l l l l

I. .I . I . i . .i . i . i . i. .i . i . i . i. .i . i . i . i. .i . i . i . .i .i i i i i i~'i::" / " i !i ': iiiiiil;iiiiiiiill

i i i i i i i I i i i i i i i i i i i i i i i i i ii~! i~i
. . . . . . . . . . . . . . . . . . . . . . . . . l

:.:.;
i2 1 3
.,...
i:i:! Total length of the proposed bridge along a l i p m e a t 1 = 68.23m
:ilil Total length of the proposed bridge along a l i p m e n t 2 8 71J1m
Total length of the proposed bridge along alignmeat 3 = 67.20m

ilii:
:!:!

Fig. 7. The plan of the bridge construction location.

the ratings assigned by the site evaluator, site 1 and site height of electric wire = 4.9 (IRC)
2 are chosen, as the difference between the rating of vertical clearance required
these two alignments is too low to discard either of above electric wires -- 0.45 m (IRC).
these at this stage.
For the present problem, the approximate total
7.2. Bridge layout planning height of the bridge from foundation level to the road
formation level is determined as 8.82 m. Hence, the
The user input and Indian Road Congress
approximate economic span range can be taken as 1.5
specifications 27 for the clearances required for the con-
to 2.0 times the total depth, i.e. 13.23-17.64m.
ceptual design process are as follows.
Therefore, the pier locations for the two selected sites,
depth of foundation viz. site 1 and site 2, are decided as described above,
(hard rock below g r o u n d ) = 2.0 m (user) and are shown in Fig. 8. Note that, in the case of site 2,
footpaths = required (user) three spans are chosen, due to the interference of
restriction on depth of deck = none (user) crossovers. Although the span lengths are slightly more
height of train above tracks = 3.2 m (IRC) than the economic range, these are adopted as the

Table 2. Classified site data and critical evaluation results

Property Site 1 Site 2 Site 3

Bridge alignment SQUARE MODERATE SQUARE

Approach alignment MEDIUM STRAIGHT CURVED
Foundation strata HARD ROCK HARD ROCK HARD ROCK
Fault .formation SAFE ZONE SAFE ZONE SAFE ZONE
Depth of strata GOOD GOOD AVERAGE
Availability of labour NEAR NEAR NEAR
Availability of material AWAY AWAY AWAY
Availability of equipment AVAILABLE AVAILABLE AVAILABLE
Overall ratings 74.46 76.71 58.71
444 H. SHIVA KUMAR et al. : KNOWLEDGE-BASED CONCRETE BRIDGE DESIGN

l abutment

I/
location

lo

/
,/
E
//
a. Bridge layout for site 1 b. Bridge layout for site 2
Fig. 8. Bridge layouts.

clearance requirements are satisfied and existing foun- are rated better than the highly rated solution at site 1,
dation strata are capable of safely transferring the viz. solution 1. A snapshot of these solutions is pre-
loads. sented in Table 4. Let the solutions chosen for prelimi-
nary design be the solution 1 of site 1, and solutions 5
7.3. Conceptual design and 9 of site 2.
Various bridge configurations for the two layouts of
the bridge shown in Fig. 8 are generated through the 7.4. Preliminary design
synthesis process. There are a total of 20 solutions The solutions that are selected on the basis of the
synthesized for the two sites. The results of a critical evaluation results in the conceptual stage are taken up
evaluation of the above conceptual designs are given in for preliminary dimensioning. The scope of the present
Table 3. From the ratings assigned to the conceptual
design solutions it is clear that the solutions (5 and 9) of
site 2 are the best among the synthesized conceptual Table 4. A snapshot of selected conceptual design details
designs. Also, notice that these two solutions at site 2 Site 1 Site 2
Solution
Table 3. Ratings of the conceptual designs details sol. 1 sol. 5 sol. 9

Overall rating Span [1] Length 15.79 23.62 23.62

Deck-type T-beam T-beam T-beam
Solution No. Site 1 Site 2 Cross-section No-bulb Bulb Bulb
Reinforcement Plain Prestressed Prestressed
1 85.107 82.241 Span [2] Length 18.07 24.26 24.26
2 81.685 76.694 Deck-type T-beam T-beam Box girder
3 78.363 68.264 Cross-section Bulb Bulb Inc-m-cell
4 74.839 68.264 Reinforcement Plain Prestressed Prestressed
5 78.262 85.124 Span [3] Length 18.07 23.62 23.62
6 74.839 76.640 Deck-type T-beam T-beam T-beam
7 71.416 76.694 Cross-section Bulb Bulb Bulb
8 67.994 76.694 Reinforcement Plain Prestressed Prestressed
9 -- 85.124 Span [4] Length 15.79
10 -- 76.694 Deck-type T-beam
11 -- 76.694 Cross-section No-bulb
12 -- 68.264 Reinforcement Plain
 H. SHIVA KUMARet al. : KNOWLEDGE-BASEDCONCRETEBRIDGE DESIGN 445

T
15o / ~
t-~- / !
10 0 ~'- ~oo-~

a) 15.79m span b) 18.07m span c) 23.62m amd 24.26m spans

i) T-beam girders for decks of different spans

~v ~ 12000
150 I I
- 4-- zoo ~ 4-200

2000 . ,., 2000 1, I1 2000 ~i

ii) Box-girder for 24.26m span

Fig. 9. Preliminarydimensionsof the mainelementsof the decksof varioussolutions.

example is limited to the preliminary design of the the solution building. Hence, all the solutions that are
superstructure only. very closely spaced with respect to their ratings should
The preliminary dimensions of the main girders be selected and developed further before eliminating
obtained from the KBS is shown in Fig. 9. These them.
solutions are evaluated on the basis of their weights by
the comparative criticism. The preliminary designs cho-
sen at this stage form the input to another computatio- 8. CONCLUSIONS
nal program for analysis, detailed design and drafting.
The preliminary designs are analysed using the avail- Design process modeling is an important step in KBS
able finite element analysis package called PASSFEM 28 development. In this paper, a process model for con-
and these results are used for detailed design of the crete road bridge design has been proposed, based on
structural sections. Since these are outside the scope of the identified AI-based techniques for the knowlege-
the present work, the details are not discussed here. intensive tasks of site selection, bridge layout planning,
The results of the comparative critical evaluation of the conceptual design and preliminary design. Based on the
preliminary design are given in Table 5. The obvious presented work, the following conclusions are arrived
at.
final choice of the KBS would be the solution 1 at site 1.
Otherwise, the user can choose all three solutions, and (1) The four tasks of the bridge design process, viz. site
make a final decision based on actual dimensions of the selection, bridge layout planning, conceptual
superstructures. design and preliminary design, require the use of
From the above example it is very clear that the following problem-solving techniques
solutions rated "high" at some stage of the solution For solution generation:
building need not be the best ones at the next stage of --production-rule-based planning in the bridge
layout planning stage
Table 5. Ratingsof the preliminarydesigns - - decomposition-based synthesis in the concep-
tual design stage
Site Solution Weightof decks
No. No. (in tons) Rating --production-rule-based generation in the pre-
liminary design stage.
1 1 700.725 100,0 For solution evaluation:
2 5 729.235 96.08
9 729.235 96.08 --critical evaluation of hierarchically repre-
sented solutions involving the determination
446 H. SHIVA KUMAR et al. : KNOWLEDGE-BASED CONCRETE BRIDGE DESIGN

of overall worth based on the ratings of indi- 7. Sause R. and Powell G. H. A design process model for computer
vidually assessed aspects a n d their relative integrated structural engineering. Engng. Comput. 6, 129-143
(1990).
i m p o r t a n c e , as in the case of site selection a n d 8. Sause R. and Powell G. H. A design process model for computer
c o n c e p t u a l design integrated structural engineering: design phases and tasks. Engng
- - c o n s t r a i n t p o s t i n g as in the case of bridge Comput. 7, 145-160 (1991).
9. Luth G. P., Jain D., Krawinkler H. and Law K. H. A formal
layout p l a n n i n g approach to automating conceptual design, Part I: methodology.
- - c o m p a r a t i v e assessment as in the case of pre- Engng Comput. 7, 79-89 (1991).
l i m i n a r y design. 10. Jain D., Krawinkler H., Law K, H. and Luth G. P. A formal
approach to automating conceptual design, Part II: application to
(2) T h e R S C m o d e l p r e s e n t e d in this p a p e r is m o r e floor framing generation. Engng Comput. 7, 91-107 (1991).
suitable for the c o n c r e t e road bridge design where 11. Maher M. L. Process models for design synthesis. AI Mag. ll,
the process is characterized by the i n c r e m e n t a l No. 4, 49-58 (1990).
12. Ahmed S., Sriram D, and Logcher R. Transaction-management
solution b u i l d i n g using r u l e - b a s e d i n f e r e n c e , syn- issues in collaborative engineering. J. Comp. Cio. Engng ASCE
thesis a n d critical e v a l u a t i o n at various stages. 6, 85-105 (1992).
(3) T h e application e x a m p l e d e m o n s t r a t e s the w o r k i n g 13. WelchJ. G. and Biswas M. Applications of expert systems in the
design of bridges. Transp, Res. Record 1072 65-70 (1986).
of the R S C m o d e l with the k n o w l e d g e compiled 14. Burgoyne C. J. and Sham S. H. R. Applications of expert
from o n e p a r t i c u l a r design practice. H o w e v e r , the systems to prestressed concrete design. Civ. Engng Syst. 4, 14-19
R S C m o d e l is g e n e r a l e n o u g h to be used for the (1987).
15. Adeli H. and Balasubramanyam K. V, A knowledge-based
d e v e l o p m e n t of a KBS for o t h e r design practices system for the design of bridge trusses. J. Comp. Cio. Engng
too. ASCE 1, No. 3 (1987).
16. Adeli H. and Balasubramanyam K. V. A Heuristic approach for
interactive analysis of bridge trusses under moving loads. Micro
Acknowledgements--This work reported in this paper is based on the Comput. Oo. Engng 2, 1-18 (1987).
lndo-US research project on "KBES for Civil Engineering Design" 17. Adeli H. and Balasubramanyam K. V. Interactive layout optimi-
under NSF Grant nos. INT-8913816 at Indian Institute of zation of trusses. J. Comp. Cir. Engng 1,183-196 (1987).
Technology, Madras in collaboration with Carnegie Mellon 18. Adeli H. and Balasubramanyam K. V. A knowledge-based
system for design of bridge trusses. J. Comp. Cio. Engng 2, No. 1
University, Pittsburgh, U.S.A. The authors express their thanks and
deep appreciation to Dr K. Srinivasa Rao, retired Superintending (1988).
19. Spences W. J., Atkins R. M. and Podlaha P. Development of an
Engineer, Highway Department, Madras, India, for providing expert
knowledge based on his long experience in the construction of expert system for the preliminary design of bridges. Cio. Engng
concrete bridges, and for his continuous interaction in the develop- Syst. 6, 51-57 (1987).
ment of this system. The authors are also thankful to Professor S. J. 20. Moore C. J. and Miles J. C. The development and verification of
a user oriented KBS for the conceptual design of bridges. Cir.
Fenves, co-investigator of the project at CMU, for introducing them
to the concept of design criticism, and Dr S. Rajeev of liT-Madras Engng Syst. 8, 81-86 (1991).
21. Choi C. K, and Choi I. H. An expert system for selecting types of
for his suggestions and help.
bridges. Computers Struct. 411, 183-192 (1993).
22, Partridge D. To add to AI or not to add to AI. Proceedings of
Expert Systems '88, Bridgeton, U.K., pp. 3-13 (1988).
REFERENCES 23. Buchanan B. G. and Shortliffe, E. H. Rule-Based Expert
Systems: The MYCIN Experiments of the Stanford Heuristic
1. Dym C. L. and Levitt E. T. Knowledge-Based Systems in Programming Project. Addison-Wesley, Reading, MA (1984).
Engineering. McGraw-Hill, New York (1991). 24. Waterman D. A. A Guide to Expert Systems. Addison-Wesley,
2. Maher M. L. Expert systems for structural design. J. Comp. Cir. Reading, MA (1986).
Engng ASCE 1,270-283 (1987). 25. Krishnamoortby C. S., Shiva Kumar H., Rajeev S. and Suaresh
3. Stefik M. J. Introductin to Knowledge Systems, Xerox Palo Alto S. Knowledge based system with generic tools for structural
Research Center, Palo Alto, CA (1990). engineering design. Struct. Engng. Reo. 5, 121-131 (1993).
4. Bedard C. and Ravi M. Knowledge-based approach to overall 26. Shiva Kumar H., Suresh S., Krishnamoorthy C. S., Fenves S. J.
configuration of multistory office buildings. J. Comp. Cio. Engng and Rajeev S. GENCRIT: a tool for knowledge-based critiquing
ASCE 5, 337-353 (1991). in engineering design. Artif. Intell. Engng Design Anal. Manuf.
5. Maher M. L. Engineering design synthesis: a domain indepen- 8, 239-259 (1994).
dent representation. Artif. Intell. Engng Design Anal. Manuf. 1, 27. IRC: 5, Standard specifications and code of practice for road
207-213 (1987). bridges, section l--general features of design (Sixth Revision).
6. Fenves S. J., Fiemming U., Hendrickson C., Maher M. L. and Indian Roads Congress (1985).
Schmitt G. IIntegrated software environment for building design 28. Krishnamoorthy C. S. Finite Element Analysis--Theory and
and construction. Computer-Aided Design 22, 27-36 (1990). Programming. Tata McGraw-Hill, New Delhi (1994).

AUTHORS' BIOGRAPHIES
H. Shiva Kumar is currently a Graduate Student in the Civil Engineering Department at the Indian Institute of Technology,
Madras. His research interests include case-based reasoning, design criticism and integrated knowledge-based systems in
general. He received his Masters Degree in Structural Engineering from Bharatiyar University, Coimbatore in 1990 and a
Bachelor of Technology degree in Civil Engineering from Jawaharlal Nehru Technological University, Hydrabad in 1988.
C. S. Krishlmmoorthy is Professor of Structural Engineering in the Department of Civil Engineering, and currently the Dean
of Academic Research at the Indian Institute of Technology, Madras. He received his Ph.D. degree from the University of
London in 1972. He obtained his BE (Hons) and [Link]. (Engineering) degrees from the Univerisity of Madras. His current
research interests include Adaptive Finite Element Analysis, Structural Optimization and Knowledge-based expert systems.
He is a fellow of the Institution of Structural Engineers, London, Fellow of the Institute of Engineers, India and Fellow of the
Indian National Academy of Engineering.
 H. SHIVA KUMAR et al. : KNOWLEDGE-BASED CONCRETE BRIDGE DESIGN 447

N. Rajagolmlan is Professor of Structural Engineering and currently the Head of the Department of Civil Engineering at the
Indian Institute of Technology, Madras. He received his Dr. Ing. degree from the University of Stuttgart in 1973, Master of
Technology degree in Structural Engineering from Indian Institute of Technology, Madras in 1966 and BE (Hons) in Civil
Engineering in 1960. His research interests include prestressed concrete bridges, finite element analysis, structural optimiza-
tion and knowledge-based systems as applied to bridge design. He is a fellow of the Indian Institute of Bridge Engineers, and a
member of the Indian Road Congress and the Indian Concrete Institute.