You are on page 1of 10

FUZZY STOCHASTIC COST AND SCHEDULE RISK ANALYSIS:

MOB CASE STUDY


Andrew Nyakaana Blair, Bilal M. Ayyub
University of Maryland at College Park1
William J. Bender
Central Washington University2

ABSTRACT feasibility study of a Mobile Offshore Base (MOB).


The MOB platform is about 1500m (1mile) by 120m
The Office of Naval Research in a feasibility study of (400feet), which is unprecedented in size and
a Mobile Offshore Base (MOB) has proposed several operations compared to any floating structure to date.
concepts for very large ocean structures and will make This paper uses the MOB hull construction as a case
risk informed assessments of alternative concepts. study to demonstrate risk-based decision analysis
Using the MOB as a case study, this paper provides a methodology.
methodology for cost and schedule risk-based decision
analysis of alternative MOB hull construction concepts. 1.2 Objective
During risk analysis, cognitive, vague or aleatory The objectives of this paper are to provide
uncertainty can be handled using fuzzy sets and logic in methodologies for the implementation of fuzzy-
many modeling scenarios. Stochastic modeling and stochastic risk-based analysis of alternative MOB hull
analysis is mandated in cases of probabilistic, construction concepts and to demonstrate fuzzy set
ambiguous or epistemic uncertainty. This paper uses a quantification of MOB subjective information. A
methodology for fuzzy-stochastic risk-based decision further objective is to discuss the use of Bayes Theorem
analysis of alternative MOB hull construction concepts for updating the state-of-knowledge of MOB cost and
and demonstrates fuzzy set quantification of MOB schedule information as it is piecewise accumulated.
subjective information. The use of Bayes Theorem for Discrete event simulation of a Critical Path Method
updating the state-of-knowledge of MOB cost and (CPM) network provides information for risk-based
schedule information as it is piecewise accumulated is decision analysis. A fuzzy logic inference system is
also discussed. used to capture both quantitative and subjective
information in the form of numerical and linguistic data
1. INTRODUCTION entry. Fuzzy based costing and scheduling enables use
of approximate MOB data. The results are incorporated
1.1 Overview in simulation program to perform fuzzy stochastic
Risk management decisions are among the most modeling. A decision tree is fed both outputs of the
conceptually difficult ones faced by managers (see [1]). fuzzy logic inference system and the network
Many projects have schedule risks inherent in their simulation. Sensitivity analysis is performed on the
execution which require appropriate modeling, analysis decision analysis results. The application of Bayes'
and assessment. Theorem to up-date information obtained from the
Risk analysis can be defined as a technique for results of the risk-based decision analysis is discussed.
identifying, characterizing, quantifying, and evaluating
hazards (see[2]). In risk analysis, the consequences 2. RISK ANALYSIS
must somehow be evaluated even if qualitatively, and
the definition of the number or quantity assigned to a 2.1 Definition of Risk
particular consequence clearly given. The concept of risk is used to assess and evaluate
Several very large ocean structures have been uncertainties associated with an event. Risk can be
proposed as part of the Office of Naval Research

1
Center for Technology and Systems Management, University of Maryland at College Park, MD 20742, USA
E-Mail: blair@eng.umd.edu and ayyub@umail.umd.edu
2
400 E 8th Avenue, Ellensburg, WA 98928, USA. E-Mail: benderw@cwu.edu
defined as the potential for loss as a result of a system uncertainty
failure. Risk can be measured as a pair of the 4. Fuzzy stochastic analysis for cases with both
probability of occurrence of an event, and the outcomes cognitive and non-cognitive types of uncertainty
or consequences associated with the event’s occurrence. where the fourth case can be developed by combining 2
This paring is not a mathematical operation, a scalar or and 3. The MOB has activities whose durations and
vector quantity but a matching of an event’s probability costs can be analyzed as deterministic, stochastic, fuzzy
of occurrence with the expected consequence. or fuzzy stochastic.

2.2 Uncertainty Modeling And Analysis 3. FUZZY STOCHASTIC RISK–BASED


DECISION ANALYSIS METHODOLOGY
2.2.1 Need for Uncertainty Treatment
A kernel element of risk is uncertainty represented 3.1 Fuzzy Stochastic Risk Analysis
by plural outcomes and their future likelihoods (see[3]). Software is available that includes automation of
There has been a paradigm shift in science concerning fuzzy set computations and interactive graphical fitting
the treatment of uncertainty. Not only is this of various distribution. In the estimation of the cost and
considered from the statistical treatment of ambiguity schedule of activities, either historical data are available
but also more applications are available, for example and may be used to categorize a statistical distribution,
fuzzy sets and logic, for the treatment of vagueness. or complete data are not available and the underlying
Vagueness is often encountered in linguistic distribution has to be based on the subjective
expressions which do not crisply place a given subject information provided by an “expert.” The second
matter in one class. As such, the degree of membership possibility is frequently encountered and can be
of each variable with in the class can be defined by modeled fuzzy stochastically (see[11]) by following
fuzzy sets and logic. It has been affirmed that most of the steps which are given in the Fig. 1.
infrastructure investment problems exhibit properties of
multiobjectivity and fuzziness in real nature (see[4]).
This nessicates fuzzy treatment of uncertainty. Subjective
Information Input

2.2.2 Treatment of Uncertainty


A review of the literature finds a variety of
treatments of uncertainty. Uncertainties can be Fuzzy Set
Quantification
classified into two broad categories, namely:
probabilistic and cognitive uncertainties (see [5]).
Others (see[6]) generally classify uncertainties Estimated Parameters
including :
concerning engineering systems as ambiguity and 1. Maximum and minimum
values
vagueness. These are also referred to these as 2. Mean and variance
epistemic and aleatory uncertainties, respectively (see
[7], [8]). Uncertainty measures can be classified into
the following three major categories (see [6]): No Is fit in graphical display of
(i) Information measures based on crisp sets and distribution satisfactory?

probability theory
(ii) Uncertainty and information measures based on Yes
fuzzy sets and possibility theory.
(iii) Uncertainty measures within the framework of Stochastic duration estimate
belief and plausibility based on the theory of
evidence.
Uncertainty can be attributed to vagueness where Input into Simulation
there are ill defined boundaries and ambiguity where module
there are several choices for a given situation (see [9]). Figure 1: Fuzzy Stochastic Application
In structural and reliability analyses the following
methods for cognitive and noncognitive types of
uncertainty are recognized (see [10]): 1. Collect and input subjective information
1. Deterministic analysis for cases without 2. Quantify subjective information using fuzzy sets
uncertainty in basic variables 3. Estimate various parameters of distributions,
2. Stochastic analysis for cases with noncognitive including maximum and minimum values, and the
type of uncertainty mean and variances of the parameters
3. Fuzzy analysis for cases with cognitive type of 4. Examine graphical display of distributions. The
fitted distribution is affected by selection and dominating distribution type. This theorem is applied
membership values of the linguistic variables. If fit to each MOB concept and scenario combination critical
is unsatisfactory, update estimated parameters. path for schedule duration and all activities for cost. In
5. From satisfactory fit obtain stochastic estimate of simulating total schedule duration, only activities on the
the duration critical path are considered while in simulating total
6. Input results into simulation module cost, all activities are considered. Schedule duration is
Quantification of subjective information using fuzzy approximated by a probability distribution with mean
sets is done using the following steps: and variance the sums of the means and variances of
Step1: Variable definition the activities on the critical path, respectively. Total
Step 2: Rule composition cost is approximated by a normal distribution with
Step 3: Translation mean and variance the sums of the means and variances
Step 4: Execution of all activities, respectively. The probability
This quantification creates the fuzzy inference system associated with completing the project within certain
depicted in Fig. 2. time or cost limits can be computed using standard
normal distribution formulae. The central limit
theorem is used as a basis to justify computed schedule
and cost risk probability.
Specify system Macro-level
structure and debugging and
variables analysis
3.2 Risk-Based Decision Analysis
Performing risk analysis may require making various
Fuzzy Inference
System
decisions in order to reduce the risk of the MOB.
Engineering decisions involving risk need to be made
using a systematic framework that considers many
facets of a decision problem. The decision framework is
Specify
membership
Micro-level
debugging and
called the decision model. In order to construct a
functions for each
variable
analysis decision model (see[14]), the following elements of the
decision model need to be defined: (1) objectives of
Specify fuzzy rules
decision analysis, (2) decision variables, (3) decision
outcomes, and (4) associated probabilities and
Figure 2: Fuzzy Inference System consequences.

3.2.1 Objectives of Decision Analysis


Linguistic variables can be translated into Engineering decision problems can be classified into
mathematical measures by fuzzy sets and systems single- and multiple-objective problems. Decision
theory and conventional procedures like the Program analysis requires the definition of these objectives. For
Evaluation and Review Technique (PERT) can be used cases of multiple objectives, the objectives need to be
if updated probabilistic input is used to obtain the stated in the same units, and weight factors that can be
required information (see[12]). The simulation module used to combine the objectives need to be assigned.
uses updated probabilistic input. Application of the
following steps results in risk estimates: 3.2.2 Decision Variables
1. CPM is performed on a project to identify critical The decision variables for the decision model need
and near critical paths also to be defined. The decision variables are the
2. Critical and near critical paths are modeled using feasible options or alternatives available to the
PERT procedures decision-maker at any stage of the decision-making
3. Continuous or discrete event simulations are run. process. Also, ranges of values that can be taken by the
The number of simulations to meet accuracy and decision variables should be defined. Assigning a value
stability requirements is determined statistically. to a decision variable means making a decision at that
4. Risk-based decision analysis is performed on point of a decision-making process. These points within
simulation results to give fuzzy stochastic risk the decision-making process are called decision nodes.
assessment.
The central limit theorem, when used in MOB 3.2.3 Decision Outcomes
construction simulation, implies that no matter what the The decision outcomes for the decision model need
underlying distribution are, the simulated schedule also to be defined. The decision outcomes are the
duration or cost mean will tend to a normal distribution events that can happen as a result of a decision. They
(see[13]) as long as the assumption of the theorem are are random in nature, and their occurrence cannot be
met of large number of input variables without a fully controlled by the decision-maker. The decision
outcomes with the associated occurrence probabilities
need to be defined. The decision outcomes can occur VI = E ( AI ) − E ( BI ) (1)
after making a decision at points within the decision-
making process called chance nodes. Where E(AI) is the utility or expected value of the
optimal alternative in the analysis after acquisition of
3.2.4 Associated Probabilities and Consequences additional information and E(BI) is utility or value
The decision variables take values that can have before acquisition of additional information. If prior to
associated costs. These costs can be considered as the obtaining information, VI exceeds the cost of
direct consequences of making these decisions. The information acquisition then the additional information
decision outcomes have both consequences and should not be acquired.
occurrence probabilities. The probabilities are needed The value of the additional information however is
due to the random (chance) nature of these outcomes. bounded by a limit referred to as the "value of perfect
information" (see[17]). The value of perfect
3.2.5 Decision Trees information (VPI) can be calculated in the using
Decision trees are systematic models that are used to Equation 1 but substituting VPI for AI. VPI represents
examine the information in decision making. Branches the maximum cost a decision maker can allow for
formed from decision or chance nodes construct a acquiring additional information.
“tree”. A decision node splits into two or more paths,
based on number of possible decisions. These paths can 3.4 Fuzzy Sets and Logic
be chosen by the decision-maker and have costs One of the ideas of fuzzy sets and logic is that
associated with them. Chance nodes split into two or artificial logic system can be developed to emulate the
more paths and follow the decision nodes. The chance linguistic way humans think and judge, yet achieve
nodes are followed by possible outcomes that can occur consistency by following accountable rules (see[18]).
without the control of the decision-maker. These The power of fuzzy set theory is that it allows a
outcomes have both probabilities and consequences formalization of vague data, a representation of their
associated with them. Each split in both decision and fuzziness that can be entered into computation and a
chance nodes form a line from left to right that is called possibility theoretic interpretation (see[19]).
a tree branch. Each branch represents a possible Fuzzy logic provides a means of performing
outcome of what can happen. Appropriate decisions linguistic computations. One of the basic ideas of
with the lowest expected costs may be chosen. fuzzy logic is that any statement employed in reasoning
will have a corresponding confidence level. Fuzzy
3.3 Bayesian Updating of Information logic also provides rules for the truth of complex
Accurate estimates of risk-based decision analysis statements. The confidence in a statement involving
parameters require substantial amount of data. When AND is the minimum of the confidences in the
observed data is limited , statistical estimates can be individual statements which make up the complex
supplemented (or even superceded) by judgemental statement. If the complex statement involves OR, the
information using the Bayesian Approach (see[15]). Its confidence in the complex statement is the maximum of
been shown that Bayes, Theorem is an effective tool for the confidences in the individual statements.
updating prior probabilities (see[16]). Fuzzy sets and logic can be used to capture
Bayes' theorem gives the rule for updating belief in a qualitative domain expert opinion for an achievable and
hypothesis (i.e. its probability) given additional affordable schedule. Although this technique cannot
evidence , and background information (context). substitute for deterministic scheduling methods, it does
Bayes theorem can be extended using the Product Rule complement the set of modeling methods thus enabling
from probability to multiple sequential updates and a better and more extensive risk assessment in cases of
applied to the conditional probabilities obtained after vague and incomplete project information.
decision analysis to update results as additional
information is collected. 4. MOB CASE STUDY
Since Bayes theorem is an effective tool for updating
prior probabilities and since subjective information can 4.1 MOB Model Building
be analyzed using fuzzy set theory, the two theories can Risk-based MOB decision analysis is performed
be combined, called fuzzy-Bayesian to compute using hierarchical modeling techniques to present
posterior probabilities (see[16]). Fuzzy-Bayesian can discrete event simulations for potential construction
similarly be used to measure the value of additional concepts and scenarios (5 five separate concepts and
information. 2 different scenarios , (see[20], [21]). One of these
The value of additional information (VI) can be concepts is used in this paper to demonstrate
measured by (see[17]): application of methodology.
Several software vendors provide programs available increase in total cost and schedule output variable is set
for both discrete and continuous event simulations. from 0% to 10%.
One such software product is Extend™ by Imagine IF-THEN rules are composed for the fuzzy inference
That, ® Inc. which is used herein. Extend is a general- engine to map input variable to output variable.
purpose graphically oriented, discrete event, and These rules are translated and executed with inputs
continuous simulation software product. for 10 to 20 shipyards. The results from this fuzzy
The model layout for the rigid concept afloat analysis are given in Figure 4 and are used in
scenario is shown in Fig. 3 . simulations.

Erect Erect Erect

Percent Increase in Total Cost


Blocks Blocks 10
Blocks Blocks
for GB1 for GB3
for GB1 for GB2
Assemble 8
Assemble

and Schedule
GB3
GB1
Blocks Blocks Assemble
Lower 6
GB2
Hulls for GB2 for GB3

4
Start 1st 2nd 3rd
Column Outfit
Column Column 2
set set set
0
1st Brace 2nd 3rd Brace 10 12 14 16 18 20
set Brace set set
Number of Shipyards in Concept and Scenario

Figure 3: Rigid Concept Afloat Scenario Model Figure 4: Fuzzy Assessment of Construction
Layout Management Conditions

The network layout in Fig. 3 is the basis for the For example, there are 12 shipyards involved in the
model produced with the Extend software. The heavy rigid afloat concept and scenario, thus the simulated
lines signify the critical path. Each MOB concept and results for cost and schedule are increased by 1.3 % to
general scenario has some common components and account for construction management conditions.
input distributions. There were a total of 9 concept
scenario combinations (see[20], [21]). Only details of 4.3 MOB Design of Experiments
one, the rigid concept with afloat assembly is given in Design of experiments for MOB construction
this paper. simulation involves the determination of the number of
simulation cycles that meet precision and accuracy
4.2 Fuzzy Assessment of Construction Management requirements. To determine the accuracy of simulation
Conditions results the number of cycles necessary to achieve an
Construction management conditions impact a given acceptable error level in the estimate of a given MOB
concept cost and schedule due to the complexity and concept and scenario schedule duration and cost are
dependence of integrating components from several computed from the results of an initial 50 cycle run. A
facilities. The uncertainty involved is subjective and standard 95% confidence level is assumed. Use is
thus incorporated into simulation using fuzzy analysis. made in this design of experiments of cost and schedule
FUZZLE 3.0 a fuzzy logic inference shell for estimates that are preliminary with a precision in the
development of rule-based system applications to 15% order of magnitude. In simulating based on this
control, diagnostics and decision aid system is used. data, a simulation accuracy (e) equal to 1% of the cost
Two variables are defined for this fuzzy analysis. and schedule of any given MOB concept and scenario
The input variable is the number of shipyard involved module is specified which is one order of magnitude
in a given concept and scenario which is a measure of more than the precision of the input data and should
the complexity involved. The output variable is the improve results.
percentage increase in total cost and schedule. With a 95% confidence level and a specified
Membership functions for the variables are accuracy (e), the minimum number of cycles (n) per
formulated by a possibility distribution function taking simulation run are given by:
values between 0 and 1. Three triangular membership σ2
functions, small, medium, and large are defined per n = 2 (1.96) 2 (2)
e
variable. The range of the number of shipyard input
variable is set from 10 to 20. The range of the percent Simulations with 50 cycles are run for each concept
and scenario combination and equation 2 is applied to
765

Mean Schedule Duration


determine the minimum number of simulations for 760
755
assumed confidence levels and specified accuracy. For 750

(Days)
the rigid concept is found that a minimum of 47 and 15 745
740
runs for schedule and cost respectively are required. 735
Thus the minimum required runs for both cost and 730
725
schedule is 47. 0 1000 2000 3000 4000 5000
Having found the minimum number of cycles Number of Simulation Cycles

required per simulation, the stability of the results from Figure 7: Mean Schedule Duration as a Function of the
simulation to simulation are investigated. Convergence Number of Simulation Cycles
tests of the coefficient of variation of the mean are done
to determine the number of cycles required to obtain
stable results. For 2000 cycles, the simulation results for all concept
The variation of coefficient of variation of the mean and scenario combinations are obtained (see[21]).
with number of cycles for the rigid concept afloat These results which apply the central limit theorem and
scenario is shown in Fig. 5. The variation of the mean use the normal distribution are used in the decision
of cost and schedule with number of cycles rigid model for decision analysis.
concept afloat scenario is shown in Fig. 6 and 7
respectively. 5. DECISION ANALYSIS

5.1 Objective of Decision Analysis


0.02 The main objectives applicable to the MOB
Coefficient of Variation Of

construction are minimizing total cost, minimizing total


0.015
schedule duration, optimizing total MOB length,
Mean

Cost
0.01
Schedule minimizing labor and safety impact, and minimizing
0.005 environmental impact. As such MOB construction
requires multiple objective decision-making. MOB
0
0 1000 2000 3000 4000 5000 objectives need to be stated in the same units, and
Number of Simulation Cycles weight factors that can be used to combine the
objectives need to be assigned. This requires
Figure 5: Coefficient of Variation of Mean Cost and
determination of objective weighting and utility for
Schedule as a Function of the Number of Simulation
multiple objective decision making.
Cycles
5.2 Objective Weighting
306 The relative weighting shown in Fig. 8 are assigned
Mean Cost (Millions $)

305
304
to the various MOB objectives.
303
302
301
300
299 Total MOB
Construction
298
297
0 1000 2000 3000 4000 5000
Number of Simulation Cycles Cost Schedule Length Labor and Safety Environmental
(35) Duration (20) Impact Impact
(25) (10) (10)

Figure 6: Mean Total Cost as a Function of the Number


of Simulation Cycles L e g e n d : Objective (weighting out of 100 total)

Figure 8: Weighting factors for MOB objectives


Figures 5 through 7 show that the simulation results
stabilize after 2000 cycles and no significant change in In setting the relative weights assigned to the various
coefficient of variation of the mean or mean cost or objectives in Fig. 8, a three step ranking procedure is
schedule is obtained with increased number of used (see [17]). First, ordinal ranking is performed,
simulation cycles. As such, 2000 cycles are adapted for whereby the 5 objectives are listed in decreasing order
use in all concept and scenario combinations. of importance. In Fig. 8, the objectives are ranked from
left to right in decreasing order of importance. Next,
the cardinal ranking of each of the objectives is
established. In this step, the relative importance of each
objective with respect to the other objectives is
evaluated by assigning numerical weights to each of the
objectives. Lastly, the weights are normalized such that 5.3.4 Labor, Safety and Environmental Impact Utility
the sum of the weights equals 100 as shown in Fig. 8. From an owner’s perspective, a MOB that has
These weight factors can also be developed using acceptable labor and safety impact and that does not
decision techniques such as the analytic hierarchy adversely impact the environment is required. As such,
process (see [22]) through pairwise comparisons of the owner has maximum utility for a MOB that meets
objectives and solving the resulting matrix of these requirements and no utility for a MOB that does
comparisons for its eigenvalues not. If the objective of labor, safety and environmental
impact are acceptable, the maximum utility is obtained.
5.3 MOB Objective Utility Functions If not acceptable, zero utility is perceived. Therefore, a
Taking into consideration the weighting factors given discrete assignment is used herein.
to the various objectives, utility functions are derived.
The utility functions represent a decision makers 5.4 Definition of Decision Variables
preference for various possible outcomes over the entire The decision variables are the feasible options or
range of possible future outcomes. alternatives currently envisaged for the MOB. These
include Concept, Scenario, Labor and Safety Impact
5.3.1 Cost Utility and Environmental Impact. For the Concept variable
The MOB is a U.S. Government funded venture and the ranges are Rigid, Hinged, Independent, Flexible, or
the Government can be assumed to be risk neutral. Concrete and Steel. For the Scenario variable the ranges
This is because the Governments subjective value of are Afloat Assembly or Terrestrial Assemble. For the
the next dollar of saving or expenditure is the same as Labor and Safety Impact variable the ranges are
the subjective value to the Government of the previous Acceptable or Unacceptable. And for the Shipyard site
dollar of saving or expenditure. The utility function is environmental impact variable the ranges are
therefore a straight line with the x-axis ranging from Manageable or Prohibitive.
$1,500 million to $6,000 million representing the
estimated range of MOB construction costs and the y- 5.5 Definition of Decision Outcomes, Associated
axis ranging from 0 to 35 utility units to conform to the Probabilities and Consequences
weights given in Fig. 8.
5.5.1 Fuzzy Assessment of Labor, Safety, and
5.3.2 Schedule Utility Environmental Impact
It is estimated that, given concepts currently under The uncertainty involved in assessing labor, safety,
consideration, it is not possible to construct an entire and environmental impacts is subjective and thus
MOB in less than 4 years. As such, maximum utility is incorporated into simulation using fuzzy analysis. Two
derived from a 4-year schedule. It is also assumed that fuzzy inference models are set up. One for labor and
any schedule greater than 12 years would have less than safety impact analysis and one for environmental
zero utility to the owner. Schedule utility between 12 impact analysis.
and 4 years is assumed to change proportionately. In In labor and safety impact analysis three variables
this utility function, the x-axis ranges from 0 to 12 are defined for fuzzy analysis. The input variables are
years and the y-axis is shown from 0 to 25 utility units the expense and the time required for a given concept
to conform with the weights given in Fig. 8. and scenario which are an indicator of the strain on the
site labor market and possibility of safety hazards. The
5.3.3 Length Utility output variable is labor and safety impact acceptability.
To meet operational requirements, MOB concepts up In environmental impact analysis three variables are
to 1500 m (1 mile) in length are being considered. defined for fuzzy analysis. The input variables are the
Various concepts have module design such that the size of the principle shipyard involved in lower hull
physical configuration of the total MOB as a construction in a given concept and scenario and the
combination of a given number of modules results in scenario itself which are measures of the requirements
total MOB length that varies with concept. In for site modification with resultant environmental
establishing length utility, it is assumed that for the impact. The output variable is an environmental impact
envisaged operational requirements, a length less than rating
2000 feet would have no utility while a MOB total Membership functions for the variables are
length of 5000 feet or greater would have maximum formulated. Triangular membership functions are
utility. In this utility function, the x-axis varies from defined and the range variables is set. For output
2000 to over 5000 feet while the y-axis is shown from 0 variables, two membership functions are defined. The
to 20 utility units to conform to the weights given in range of the impact ratings given by the output
Fig. 8. variables is set from 0 to 1. Rules, composed for the
fuzzy inference engine to map input variable to output Ingalls Shipyard
Figure 11 Total Expected Utility
variable, are developed. Six rules are developed for Alabama Shipyard
Similar to Figure 11 Total Expected Utility
both labor and safety impact analysis and five Afloat Assembly
Cost Utility = 33 Avondale Shipyard
environmental impact analysis. For example, one of Schedule Utility = 8
Similar to Figure 11 Total Expected Utility

the rules developed for environmental impact analysis Newport News Shipyard
Similar to Figure 11 Total Expected Utility

is: Kvaerner Shipyard


Similar to Figure 11 Total Expected Utility
RULE 1:IF { Shipyard Size is tiny OR Shipyard Size NASSCO
Similar to Figure 11 Total Expected Utility
is very small OR Shipyard Size is small } AND
Concept is Terrestrial THEN Environmental Impact is Newport News Shipyard
Similar to Figure 11 Total Expected Utility
Terrestrial Assembly
prohibitive Cost Utility = 32
Schedule Utility = 6

Figure 10: Portion of Decision Tree for Rigid Concept


5.5.2 Fuzzy Translation and Utility Mapping
Scenario and Shipyard Selection
These rules are translated and executed and the
possibility given by the fuzzy inference engine for the
occurrence of a given output is obtained. This Manageable Environmental Impact
possibility represents a subjective degree of confidence Acceptable Labor and Safety Impact P=0.8 , U=10
in the truth of the proposition that a given outcome will P=0.6 , U=10 P=0.2 , U=0
occur. The possibility value is a subjective Prohibitive Environmental Impact
interpretation of probability. The possibility value is Manageable Environmental Impact
normalized and is used as the probability in the decision P=0.4 , U=0 P=0.8 , U=10
trees. Unacceptable Labor and Safety Impact P=0.2 , U=0
The cost and schedule results obtained from concept Prohibitive Environmental Impact
and scenario simulations are mapped to the respective
Figure 11: Portion of Decision Tree for Rigid Concept
utility function to obtain the utility of the results. The
Labor, Safety and Environmental Impact Outcome for
total lengths of the various MOB concepts are mapped
Ingalls Shipyard
to the length utility function to obtain the utility of each
concepts total length (see [21]). The utilities are
incorporated into decision trees.
Fig. 9 through Fig. 11 are used to illustrate how the
total expected utility is computed. The expected branch
5.6 Decision Trees
utility is computed for the end of the right-most
Selected parts of the decision tree for the MOB are
decision node in Fig. 11. At decision nodes, the branch
depicted in Fig. 9 through Fig. 11. The decision nodes
with the highest utility is chosen while at chance nodes,
are identified in the model using a square symbol and
the expected utility is computed.
the chance nodes are identified with a circle symbol.
For example, the total expected utility for the Ingalls
Shipyard, utilized for afloat assembly of the rigid
Rigid Concept : Length Utility = 12 concept is obtained as follows:
Figure 10
Length Utility = 12 (From Fig. 9)
Hinged Concept : Length Utility = 20
Similar to Figure 10 Cost Utility = 33 (From Fig. 10)
Independent Concept : Length Utility = 19
Schedule Utility = 8 (From Fig. 10)
Similar to Figure 10 Labor and Safety Impact Utility = 6 (From Fig. 11)
Environmental Impact Utility = 16 (From Fig. 11)
Similar to Figure 10
Flexible Concept : Length Utility = 20 Total Expected Branch Utility = 75
Similar to Figure 10 This procedure is used to compute expected branch
Concrete and Steel Concept: Length Utility = 15
utilities for the entire MOB (see[21]).
Figure 9: Decision Tree for MOB Concept and
Scenario Selection 5.7 Sensitivity Analysis
The utility values for the various concepts and
scenarios as obtained from the decision trees are based
Only a portion of the decision tree required to on various parameters and assumptions, which are
represent the information for the MOB in this report is subject to change. One such parameter is the total
shown in Fig. 9 through Fig. 11. The entire decision MOB length necessary to meet operational
tree (see[21]) consists of Fig. 9, 5 figures similar to Fig. requirements. It is assumed that a MOB length of 5000
10 (one for each concept) and 25 figures similar to Fig. feet or greater has maximum length utility. A
11 (one for each shipyard in each concept). sensitivity analysis is done to examine the change of
concept and scenario total utility with a different
maximum utility MOB overall length. Due to changes When additional information is obtained, the
in scope of MOB operational requirements, a total probabilities used in the decision trees obtained from
MOB length of 6000 feet or greater may be appropriate. simulation and fuzzy analysis can serve as prior
Under this scenario, in establishing length utility, it is probabilities. These probabilities can then be updated
assumed that for the envisaged operational using Bayes' Theorem to obtain posterior probabilities.
requirements, a length less than 3000 feet would have In cases where some of the probabilities are fuzzy,
no utility while a MOB total length of 6000 feet or fuzzy-Bayesian updating is applied.
greater would have maximum utility.
In the resultant straight line utility function, the x- 6. CONCLUSIONS
axis varies from 3000 to over 6000 feet while the y-axis
is shown from 0 to 20 utility units to conform with the A fuzzy logic inference system is used to capture
weights given in Fig. 8. both quantitative and subjective information. This
A sensitivity analysis on these results with respect to information is used in discrete event simulation
increasing the entire MOB length to 6,000 feet yields networks to perform fuzzy stochastic modeling. A
the results given in Table 1 (see [21]). decision tree is fed both outputs of the fuzzy logic
inference system and the network simulation.
The decision analysis process combines the potential
Table 1. Sensitivity Analysis Results effects of cost, schedule, length, labor, safety and
Concept Scenario Utility Utility environmental impact to produce utilities to guide
(type of with 5000 with 5000 decision-makers. The sensitivity analysis results
assembly) foot MOB foot MOB demonstrate the effect on utility of specified changes in
required required selected inputs. Other variations may be examined as
Rigid Afloat 72 43 required by the decision-maker. For example the
Terrestrial 69 43 decision-maker may require that the results of the
Hinged Afloat 60 59 decision model be analyzed for sensitivity to variation
Terrestrial 62 61 in choice of cost and schedule probability of occurrence
Independent Afloat 66 62 and objective weighting. Not only could the factors
given be varied but also additional factors can be taken
Flexible Afloat 62 49
into account. For example, the decision-maker could
Bridge Terrestrial 63 51 request inclusion of an additional factor such as
Steel & Afloat 65 65 shipyard availability since the larger shipyards have
Concrete Terrestrial 48 46 greater utility but may not be available.
At the decision-makers discretion further
investigation may be done to obtain more discerning
This analysis is based on mean simulation results. results. Acquisition of further information has to
Table 1 shows that the rigid afloat concept has the balance investigation costs with the value of additional
greatest utility when the required MOB length is 5,000 of information. Any new information can be used to
feet while the steel and concrete afloat assembly has the update the risk-based decision analysis using Bayesian
greatest utility when the required MOB length is 6,000 or fuzzy-Bayesian updating.
feet. The independent, steel and concrete afloat Methodologies for the implementation of fuzzy-
assembly and hinged terrestrial assembly concept and stochastic risk-based analysis of alternative MOB hull
scenario total utilities are found to be insensitive to construction concepts are given. Fuzzy set
required length variation within the 5,000 to 6,000 foot quantification of MOB subjective information is
range. demonstrated. The use of Bayes Theorem for updating
The variation from shipyard to shipyard and from the state-of-knowledge of MOB cost and schedule
concept to concept is not large. The highest ranked and information as it is piecewise accumulated is discussed.
the next highest ranked concept, scenario and shipyard
have less than a 5% difference in utilities. This is an References
indication of the need for further investigation to obtain
more discerning results. Acquisition of further [1] R. J. Engemann,, H. E. Miller and R. M. Yager.
information has to balance investigation costs with the “A Decision Making Model Under Uncertainty
value of sample of information. The value of the Applied to Managing Risk” Uncertainty Modeling
additional information however is bounded by a limit and Analysis, Theory and Applications, Ayyub,
referred to as the value of perfect information as B.M. and Gupta, M.M., Editors , Elsevier, New
previously described. York, pp. 221-232, 1994.
[2] M. Modarres. What Every Engineer Should Know Applications, Ayyub, B.M. and Gupta, M.M.,
About Reliability and Risk Analysis, Marcel Editors , Elsevier, New York, pp. 287-304, 1994.
Dekker, Inc., New York, Basel, Hong Kong, 1993.
[12] B. M. Ayyub and A. Haldar. “Project Scheduling
[3] H. Kumamoto and E. J. Henley. Probablistic Risk Using Fuzzy Set Concepts” J. Constr. Engrg. and
Assessment and Management for Engineers and Mgmt., ASCE Vol.110 No.2, pages 189-204, 1984.
scientists, IEEE Press, New York, 1996.
[13] G. Smith. Statistical Reasoning, Allyn and Bacon,
[4] G.-H. Teng and J.-Y. Teng. "Multiobjective Needham Heights, Massachusetts, 3rd edition, 1991.
decision making with fuzzy problems for
infrastructure investment planning; a case of [14] B. M. Ayyub and R. H. McCuen. Probability,
transportation investment” Analysis and Statistics and Reliability for Engineers, CRC Press,
Management of Uncertainty, Theory and Washington, D.C, 1997.
Applications, Ayyub, B.M., Gupta, M.M., and L.N.
Kanal Editors , Elsevier, New York, pp. 303-318, [15] A. H.-S. Ang and W. H. Tang. Probability
1992. Concepts in Engineering Planning and Design,
Volume I Basic Principles, John Wiley and Sons,
New York, 1975.
[5] M. M. Gupta. “Intelligence, uncertainty and
information” Analysis and Management of [16] K. C. Chou and J. Yuan. Fuzzy-Bayesian
Uncertainty, Theory and Applications, Ayyub, Approach to Reliability of existing Structures"
B.M., Gupta, M.M., and L.N. Kanal Editors , Journal of Structural Engineering, Vol. 119, No.
Elsevier, New York, pp. 3-12, 1992. 11, pages 3276-3290, 1993.

[17] A. H.-S. Ang and W. H. Tang. Probability


[6] A. Ibrahim and B. M. Ayyub. “Uncertainties in Concepts in Engineering Planning and Design,
Risk-Based Inspection of Complex Systems” Volume II Decision, Risk and Reliability, John
Analysis and Management of Uncertainty, Theory Wiley and Sons, New York, 1984.
and Applications, Ayyub, B.M., Gupta, M.M., and
L.N. Kanal Editors , Elsevier, New York, 247-262. [18] L. Chao and M. J. Skibniewski. “Fuzzy Logic for
Evaluating Alternative Construction Technology” J.
[7] E. M. Pate’-Cornell. “Uncertainties in Risk Constr. Engrg. and Mgmt., ASCE Vol. 124, No. 4,
analysis: Six levels of treatment” Reliability pages 297-304, 1998.
Engineering and System Safety, Vol. 54, pages 95-
111, 1996. [19] T. Fetz, M. Oberguggenberger, J. Jager, D. Koll, G.
Krenn, H. Lessman and R. F. Stark. “Fuzzy Models
[8] E. M. Pate’-Cornell. “Global Risk Management” in Geotechnical Engineering and Construction
Journal of Risk and Uncertainty, Vol.12, pages 239- Management” Computer-Aided Civil and
255, 1996. Infrastructure Engineering, Vol. 14, pages 93-106,
1999.
[9] M. H. M. Hassan and B. M. Ayyub. “Multi-
attribute Fuzzy Control of Construction Activities [20] W. J. Bender, B. M. Ayyub, and A. N. Blair.
Uncertainty Modelling and Analysis, Theory and Assessment Of The Construction Feasibility Of The
Applications, Ayyub, B.M. and Gupta, M.M. Mobile Offshore Base. To appear in Proce. Int.
Editors , Elsevier, New York, pp. 271-285, 1994. workshop on Very Large Floating Structures, VLFS
’99, Honolulu, HI, 1999.
[10] B. M. Ayyub and R.-J. Chao. “Uncertainty
Modeling in Civil Engineering with Structural and [21] B. M. Ayyub, W. J. Bender, and A. N. Blair.
Reliability Applications,” Uncertainty Modeling "Assessment Of The Construction Feasibility Of
and Analysis in Civil Engineering, Ayyub, B.M. The Mobile Offshore Base: Part III - Construction
Editor , CRC Press, Washington, D.C., pp. 3-32, Risk Analysis," Office of Naval Research and Naval
1998. Facilities Engineering Service Center, Code ESC
51, 1100 23RD Avenue, Port Hueneme, CA, 1999
[11] S. M. AbouRizk, D. W. Halpin and A. Sawhney.
“Modeling Uncertainty in Construction Simulation" [22] T. L. Saaty. The Analytic Hierarchy Process. RWS
Uncertainty Modeling and Analysis, Theory and Publications, Pittsburgh, PA, 1996.

You might also like