See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/27523248

A Stochastic Approach to Multi-disciplinary Aircraft Analysis and Design

Article · January 1998

DOI: 10.2514/6.1998-912 · Source: OAI

CITATIONS READS
89 482

4 authors, including:

Dimitri N. Mavris
Georgia Institute of Technology
1,570 PUBLICATIONS 13,882 CITATIONS

SEE PROFILE

All content following this page was uploaded by Dimitri N. Mavris on 18 December 2014.

The user has requested enhancement of the downloaded file.

AIAA 98-0912
A Stochastic Approach to Multi-disciplinary
Aircraft Analysis and Design
D. N. Mavris, D. A. DeLaurentis, O. Bandte, M. A. Hale
Georgia Institute of Technology
Atlanta, GA
 A Stochastic Approach to Multi-disciplinary Aircraft Analysis and Design

Dimitri N. Mavris , Daniel A. DeLaurentis à, Oliver Bandte à, Mark A. Hale ¤

Aerospace System Design Laboratory (ASDL)

School of Aerospace Engineering
Georgia Institute of Technology
Atlanta, GA 30332-0150

Abstract Definitions

Within the context of multi-disciplinary aircraft Since many of the topics discussed in this paper
analysis and design, a new approach has been formulated represent concepts with which the reader may not be
and described which allows for the rapid technical familiar, a few key definitions are offered for clarity:
feasibility and economic viability assessment of multi- Ambiguity: The un-described and vague (linguistically)
attribute, multi-constrained designs. The approach, portion of a design [1]. Ambiguity occupies the
referred to here as Virtual Stochastic Life Cycle Design, space complement to knowledge.
facilitates the multi-disciplinary consideration of a Conflict: Conflict occurs when an objective cannot be
system, accounting for life-cycle issues in a stochastic extremized to the greatest possible degree since
fashion. The life-cycle consideration is deemed essential such a strategy would cause other effects that would
in order to evaluate the emerging, all encompassing result in a degradation of the objective [2].
system objective of affordability. The stochastic Decision Maker: Someone (a professional), or a team of
treatment is employed to account for the knowledge professionals, who has authority to allocate
variation/uncertainty that occurs in time through the resources and has responsibility for the output
various phases of design. Variability found in the decision.
treatment of assumptions, ambiguous requirements, Decision Making: An intelligent activity aimed at
code fidelity (imprecision), economic uncertainty, and allocating resources in order to develop a system to
technological risk are all examples of categories of meet the customerÕs expectations and requirements.
uncertainty that the proposed probabilistic approach can Decision Support: A methodological and technical
assess. For cases where the problem is over-constrained environment which facilitates the decision making
and a feasible solution is not possible, the proposed process.
method facilitates the identification and provides Fast Probability Integration (FPI) [ 3, 4, 5]: A family
guidance in the determination of potential barriers which of probabilistic analysis techniques characterized by
will have to be overcome via the infusion of new better efficiency and transparency rather than Òbrute
technologies. The specific task of examining system forceÓ probabilistic techniques such as the Monte
feasibility and viability is encapsulated and outlined in a Carlo (MC) Simulation.
series of easy to follow steps. Finally, the method Feasible Alternative: A design alternative which
concludes with a brief description and discussion of satisfies all imposed constraints (i.e. it is
proposed decision making techniques to achieve optimal physically realizable).
designs with reduced variability. This decision making Metamodel: An approximation of a complex analysis
is achieved through a combined utility theory and model. Typical metamodels include regression
Robust Design Simulation approach. models of complex computer programs based on
experimental designs (e.g., the Response Surface
Method), artificial neural networks, fuzzy sets, or
Asst. Professor, School of AE, AIAA Senior Member other metamodel building methods [ 6, 7].
à
Graduate Research Asst., AIAA Student Member Metric: A Figure of Merit that characterizes a discipline
¤
Research Engineer II, AIAA Professional Member or function or their related technologies (e.g., L/D
Paper presented at the 36th Aerospace Sciences for aerodynamics or SFC for propulsion).
Meeting & Exhibit, Reno, NV, January 12-15, 1998 Probabilistic Analysis: Analysis which allows for the
examination of systems with imprecise or
Copyright Ó 1998 by the American Institute of
incomplete information (i.e., uncertainty and
Aeronautics and Astronautics, Inc. All rights reserved.
ambiguity). In other words, a means of forming

1
American Institute of Aeronautics and Astronautics
 relationships between input and output variables, offs in decision making between the various attributes
including the variability of the inputs. that comprise operational effectiveness. A designer
Risk: Risk can be defined as the probability or chance must further distill information about these potential
of achieving an unfavorable outcome. solutions using ambiguous product requirements
Robust Design: A design which is least sensitive to definitions, incomplete data models, and under the
influence of uncontrollable factors. A solution that pressures of time and budget constraints. In most cases,
optimizes affordability while reducing associated cost must be reduced or kept under control without
variability. degrading the effectiveness of the system. In order for
Stochastic Process: Uncertain history of response over this to be achieved, the designer must gain a clear
the range of time values. insight as to the impact that his/her decisions have on
Subjective probability: A probability which has no the various attributes and the associated cost. In
specific definition but is based on experience, addition, due to the reduced budgets and number of new
expert opinion, intuition, or educated guesses. designs studied, the opportunity for expensive flight test
Uncertainty: An estimate of the difference between programs is also reduced. Hence, the design must be
models and reality. Uncertainty is manifested when able to accomplish these without the resource of
quantities associated with the product can not be historical databases. This ÒdilemmaÓ is perhaps best
determined exactly, and is a term describing the illustrated by Figure 1. For new complex systems, the
imprecision in establishing the value of a variable. design team is asked to make decisions in the early
Viable Alternative: A design alternative which is phases of the process, with relatively minimal
feasible and meets or exceeds the customer knowledge which constrain the configuration, reduce the
objective(s) (i.e., it is physically realizable and design freedom, and greatly affect the costs committed.
affordable). The variation of design knowledge, design freedom,
Virtual Design: Assessment of real-time interactive and cost commitment as a program proceeds through the
computer simulation of physical interactions in various design phases are depicted in Figure 1.
engineering systems. Inspection of this figure shows that design freedom
rapidly decreases, while the knowledge about design is
Introduction slowly increasing, and that cost commitment (life cycle)
gets locked in early. This is particularly true for
The engineering design community is presently in complex engineering systems. To remedy this
the midst of a paradigm shift. Recent initiatives in situation, a design process is desired which brings more
government and industry, focused on system knowledge to the earlier product development design
affordability as the overall decision making objective, phases, where leverage is greatest, keeping the design
are defining and encouraging this shift, and have freedom open longer; and shifting to a more gradual cost
provided the motivation and framework for the research commitment curve, ideally following the trend of how
presented in this paper. cost is expended.
The selection of affordability as the design driver The authorsÕ current ideas on how to facilitate this
denotes a dramatic change in the mindset of how paradigm shift from deterministic, performance based
complex systems are designed and built today. ÒDesign multi-disciplinary design to a stochastic formulation
for AffordabilityÓ implies that the design and evaluation whose goal is maximizing affordability are the focus of
of a system is no longer dictated solely by mission this paper. The framework for a new stochastic design
capability requirements, or even product characteristics. methodology which accounts for uncertainty, and
Instead, it is a robust decision making design process incorporates physics-based disciplinary analysis, has
that balances mission capability with other system been formulated and is presented here. This approach is
effectiveness attributes, while keeping cost under close characterized by its use of metamodels to generate
attention. This balance between benefit and cost is the higher fidelity, physics (or process) based information
main foundation of Design for Affordability, and it may to pass on to a sizing and synthesis tool that has the
be viewed simply as a measure of value, represented as central role of the integrator in the multi-disciplinary
the ratio of benefits provided or gained from the product design formulation. The incorporation of vehicle
or service to the cost of giving or achieving those economic and operational dynamics, combined with a
benefits. time varying probabilistic algorithm, and the
In addressing Design for Affordability, the designer employment of advanced decision making techniques in
must develop the system by accounting, from the searching for affordable designs, complement this
outset, for its life-cycle behavior, and allowing for trade- physics-based sizing and synthesis tool to form Virtual
Stochastic Life Cycle Design (VSLCD).

2
American Institute of Aeronautics and Astronautics
 TodayÕs Design Process
Future Design Process Weapon System Effectiveness- Aircraft Example
100%
Cost Operational Effectiveness

¥ Acquisition cost Survivability Readiness Capability Dependability

¥ Operation cost ¥ Safety

st
¥ Susceptibility ¥ Maintainability ¥ Performance
Knowledge About Design, Design

Co
Freedom, and Cost Committed

¥ Maintenance cost ¥ Reliability

¥ Vulnerability ¥ Inherent availability ¥ Maneuverability

¥ Aircraft replacement ¥ Maintenance

¥ Reliability ¥ Satisfying mission
defects
requirements
¥ Crew replacement
dge

¥ Logistics support ¥ Design defects

dge
training
¥ RDT&E Cost ¥ Operations
le

wle
Know

Kno

50%

st Figure 2: Weapons System Effectiveness

Free
Co

[Adapted from Ref. 8]

m do supportability. The cost associated with achieving

this effectiveness may be defined to be the acquisition,
Freedo
m or procurement cost, RDT&E, etc., as depicted in
0% Figure 2, or the all encompassing Life-Cycle Cost.
Preliminary
Conceptual

Detailed Manufacture Service Retirement

Life-Cycle Multi-Disciplinary Design
Multiple Concepts
As discussed previously, budget requirements
Characteristic of

Bill of Materials
Low Fidelity
Complete Product Definition
Rapid Assessments
have forced a paradigm shift from design for
Phase

Operations
Trade-off Analysis
Marketing
Medium Fidelity Environmental Impact performance to design for affordability. This shift
Sub-Component Validation Customer Satisfaction
High Fidelity
calls for new, revolutionary concepts outside the
Performance Engineering
Completed
traditional, historical databases, and demands the
consideration of all life-cycle associated implications.
The life-cycle of a product can be defined by a number
Figure 1: Life-Cycle Design Stages
of discrete phases through which the product proceeds
from concept formulation to retirement. For example,
Critical Issues Associated with Affordability
an aircraft system, similar to any other complex
engineering systems, undergoes the phases of
Affordability does not imply low cost, instead it is Conceptual, Preliminary, Detailed, Manufacture,
a measure of a systemÕs overall effectiveness which Service, and Retirement. Engineering design deals
calls for a balance between a systemÕs effectiveness and explicitly with the Conceptual through Manufacturing
the operational cost associated to provide those benefits. phases while respecting complete life-cycle
As an example, the attributes of a military aircraft implications. Each phase has a considerable impact on
system may be categorized as illustrated in Figure 2. the product as in Figure 1 [9]. It is evident, though,
With this representation of the attributes in mind, that the most leverage may be found during the early
an inclusive metric for system effectiveness is its phases of design. Making educated decisions (increased
Affordability and may be defined as the ratio of: knowledge) early on, and maintaining the ability to
Affordability =
Operational Effectiveness (1) carry along a family of alternatives (design freedom) is
Cost of Achieving This Effectiveness the key to success for the aforementioned paradigm
In order to identify the disciplines/sciences needed shift. The use of modeling and simulation, of course,
to measure and predict affordability, one must examine is a prime example of a way to shift knowledge forward
all of the key attributes which contribute to system and to capture parametric definitions of the design space.
effectiveness. Therefore, system effectiveness can be In many ways, this paradigm shift may be viewed
formally defined by: as a natural extension of the ongoing research conducted
System Effectiveness = k1(Capability) +k2(Survivability)
in the field of Multi-disciplinary Analysis and Design
+k3(Readiness)+k4(Dependability) (2)
Optimization (MDA/MDO). As the term implies,
The metric coefficients, k i , provide the ability to MDA/MDO deals with the analysis and optimization
tailor this effectiveness to specific needs, preferences, or process of multi-disciplinary problems. In this
points of view of a customer. These attributes are formulation, an objective is identified, subject to a set
directly linked to the traditional product and process of constraints, along with a set of design variables,
disciplines such as aerodynamics, structures, which are varied so as to yield an optimum solution.
propulsion, signatures, manufacturing, and Since these elementary parameters often arise from

3
American Institute of Aeronautics and Astronautics
 COMPUTER-INTEGRATED
ENVIRONMENT
QUALITY ENGINEERING TOP-DOWN DESIGN SYSTEMS ENGINEERING
METHODS DECISION SUPPORT PROCESS METHODS

7 M&P TOOLS AND ESTABLISH REQUIREMENTS

QUALITY FUNCTION THE NEED & FUNCTIONAL
DEPLOYMENT (QFD) ANALYSIS

PRODUCT DESIGN DRIVEN

PROCESS DESIGN DRIVEN
DEFINE THE SYSTEM DECOMPOSITION
PROBLEM &
FUNCTIONAL ALLOCATION

ESTABLISH VALUE
OBJECTIVES

ROBUST DESIGN GENERATE FEASIBLE

ALTERNATIVES SYSTEM SYNTHESIS
ASSESSMENT &
THROUGH MDO
OPTIMIZATION
EVALUATE
ALTERNATIVES

ON-LINE QUALITY
SYSTEMS ANALYSIS
ENGINEERING & MAKE DECISION &
STATISTICAL PROCESS
CONTROL
CONTROL (SPC)

Figure 4: Key Elements Needed for IPPD

Figure 3: IPPD and MDO

requirements may be stated ambiguously, especially in
the initial development stages. Proper representation 100%

requires the identification, understanding, modeling, and

translation of life-cycle customer requirements including Ambiguity

Knowledge About a Product

market considerations to the design functions.
The un-described and vague (linguistically) portion
of a design is the ambiguity that is present [13].
50%
Ambiguity occupies the space complement to
knowledge as in Figure 5. Uncertainty arises because Knowledge
quantities associated with the product cannot be
determined exactly and the knowledge curve boundary is
Uncertainty
unknown.
Emphasis is placed here on measuring, quantifying, 0% Time
and integrating customer and market inputs with Demonstration Production &
& Deployment
technological considerations to develop innovative Validation
Concept Engineering & Operation
technologies, and design optimal products. Techniques Exploration & Manufacturing &
Definition Development Support
such as Natural Language Processing, intelligent
modeling and control through soft computing, Figure 5: Uncertainty Variation in Time
possibility models, and conflict identification methods distribution. This variable distinction follows
are some of the approaches proposed in literature. TaguchiÕs definitions [15]. Means are needed to
analytically quantify and control design uncertainty in
Model Fidelity Representation multi-objective design problems to yield robust designs.
A multi-disciplinary treatment of design for This can be facilitated by emerging techniques from
affordability calls upon the integration of various mathematics and soft computing disciplines.
analytical methods (implemented as computer codes) at
different stages of the design life-cycle. The fidelity of Stochastic Nature
these codes is generally not equal nor known. Another
The predicted forecast response of some Overall
form is introduced as a consequence of inadequate
Evaluation Criterion (OEC) is depicted in Figure 6. As
analytical models or the dynamic nature of a system
indicated in the figure, during the conceptual phase, the
(such as the evolution of a design as it progresses from
distribution associated with such an OEC has
conceptual to detailed design). Even in the best of
significant variability but barely meets the target. As
circumstances, the uncertainty associated with these
the design process progresses in time through the
estimates is not well known. In such cases, the
preliminary and detailed phases, the knowledge about
statistics are unknown. Therefore, fidelity must be
the design progressively increases. This can be seen by
determined along with relationships which link the error
the shrinking variability. Furthermore, a shift of this
to operating conditions.
distribution closer to the target is desirable and pursued.
This time or design phase dependency with uncertainty
Design and Operational Uncertainty
dictates the need for a stochastic treatment.
Design uncertainty is an inability to analytically Early in the program, the predicted OEC estimate is
predict the outcome of an event, or the exact value of a well removed from the selected target and has skewed
parameter. Operational uncertainty arises as a result of probability distribution. As the program becomes
what are often called noise parameters that affect the better refined and the OEC estimate shifts closer toward
performance of a system. Hence, two distinct classes of the cost target, the probability distribution shifts to
design parameters emerge: control parameters and noise more of a normal distribution. This example has a
parameters. Control parameters are items that the close analogy to process capability indices, C p and C pk,
designer has direct control over, while noise parameters used in on-line manufacturing to reduce defects. Hence,
are items that effect the design, but are beyond the on-line robust manufacturing techniques, such as the
control of the designer. Hence, they should not be set use of process capability indices, also have a place in
to a single point value. Instead, they should be off-line manufacturing, i.e. the design phase.
specified in terms of a range and probability

5
American Institute of Aeronautics and Astronautics
 the knowledge available about a product at the time the
Upper Specification decision is to be made. This corresponds to the leftward
shift in the knowledge vs. freedom curve in Figure 1.
The central element in the proposed method is a
framework for modeling aerospace systems in a
Lower Specification OEC Target stochastic fashion, adhering to the following principles:
Time physics-based analysis with associated metamodels are
Initial Distribution Reduced Variability and Improved Mean Response needed to replace relationships based on historical
Conceptual Preliminary Detail databases (which are likely to be obsolete for current and
Figure 6: Variability of Design in Time future vehicles and subsystems). The behavior of the
systemÕs entire life-cycle must be represented in
synthesis and optimization models, and uncertainty
Multiple Attribute Decision Making must be incorporated and mitigated. This framework is
Multiple attribute decision making (MADM) refers shown in Figure 7 and is called Virtual Stochastic Life-
to making decisions in the presence of multiple, usually Cycle Design (VSLCD). Traditional aerospace design
conflicting, criteria. For the case of design for frameworks often stop after synthesis/sizing and
affordability, the decision maker is asked to trade-off optimization. This practice is unacceptable in the
survivability, capability, dependability, and readiness. emerging paradigm where non-deterministic models and
for a variety of scenarios. For instance, optimize the objectives are potential sources of system variability
aforementioned attributes, while minimizing or keeping which can affect design decisions. VSLCD addresses
cost fixed at a given level to obtain a configuration that this problem by incorporating all phases of design via a
satisfies minimum acceptance levels. Another scenario virtual life-cycle model.
includes keeping cost fixed while performing trade-offs The purpose of VSLCD is to facilitate decision
for the various attributes so as to obtain an optimal making (at any level of organization) to reach affordable
level of the OEC. This is performed by placing conclusions with adequate confidence. The ÒVLCÓ in
subjective weight factors in front of each attribute (Eq. VSLCD implies that it eventually will encompass in a
2). For cases where uncertainty is included, the process virtual manner the entire life-cycle including design,
must also consider solutions or decisions that are engineering development and testing, manufacturing,
robust, i.e., compromised solutions that reduce the flight test, and an operational simulation (which will
associated variability. The distinguishing feature of include certification, testing and evaluation, fielding of a
MADM is then to select the best of a finite number of vehicle in the existing infrastructure, and tracking of its
design solution alternatives. These alternatives have an impact on the economy, market demands, environment,
associated level of achievement for the attributes based etc.). The word stochastic has been added to VSLCD to
on which the final decision is to be made. The final indicate that, in the presence of possibly time-varying
selection of the alternative is made with the help of uncertainty, the method will define, mitigate, and
inter- and intra-attribute comparisons. The comparisons control variability via stochastic methods (e.g.
may involve explicit or implicit tradeoffs [16]. probabilistic, fuzzy, etc.). This capability will enable
the designer to assess a design with a high degree of
New Methodology Formulation confidence.
VSLCD deals with the processes of using analyses
Design, in the context of this paper, can no longer in sizing, synthesis, mission simulation, and eventually
be viewed as a deterministic process. In fact, a in assisting a designer in making decisions. As
probabilistic approach is needed where ranges and shapes mentioned previously, these tasks are complicated by
for all contributing inputs are available either elements such as ambiguous customer requirements,
objectively, when the statistics are known or operational uncertainty, and technology risk. A primary
subjectively, Òfuzzy probabilisticsÓ, when data is goal is to understand the nature and variety of design
unavailable and ranges are determined based on expert uncertainty and to find ways to analytically quantify and
opinion. Realizing that uncertainty varies with time, as control uncertainty in multi-objective design problems.
knowledge increases about the design (Figure 5), it A VSLCD capability will enable a designer to
becomes evident that a time varying probabilistic assess a design with a corresponding confidence
problem needs stochastic treatment. estimate. Customer requirements are translated into
A key assumption in the decision making paradigm metrics that are better defined in engineering terms, and
is that a designer will make the ÒbestÓ decisions with they may change during the design/development
process. Operational/environmental uncertainty relates

6
American Institute of Aeronautics and Astronautics
Figure 7: Virtual Stochastic Life-Cycle Design (VSLCD)
synthesis will allow life-cycle disciplines such as a design variable, the accuracy of an objective function
economics and support to be addressed. value, modeled by computer simulation or any other
Decision Support: Utility is a high-level metric engineering model, can also be quite uncertain too.
that allows a designer to measure progress and efficacy This type of uncertainty is referred to as fidelity, and is
of candidate solutions. A decision support environment a typical problem in models based on historical data.
provides design guidance as design alternatives are Fidelity can be modeled with an error term e that is
explored along alternate decision paths. added to the objective function value [20]. e is a
Integration: Simulation is critical to the random variable with a standard normal distribution. A
determination of product characteristics. Agent third type of uncertainty arises through the inclusion of
technologies and metamodeling facilitate the variety of information about the readiness of new technological
analysis methods. In addition, advanced data structures concepts and their associated risk into the design
being developed that allow stochastic parameter process. This information can also be modeled with
information to be tracked in addition to traditional random variables by recognizing the uncertain value of
deterministic values. the metric in question, and assigning an appropriate
Decision Making: Feasible and viable designs are distribution to that metric [21]. The result of the
determined through a five step process. Robust Design analysis in all three cases will be a probability
Simulation will yield robust concepts. Probabilistic distribution for the objective function. This
methods are employed to capture the influence of code distribution is then used in the robust design evaluation.
fidelity, operating uncertainty, and requirement It is noteworthy here, that the information
ambiguity. Finally, utility theory is used to facilitate modeling described above is somewhat tailored to the
multi-attribute decision making, used to weight physics based modeling. If no such analysis tools are
potentially conflicting requirements. available, more generic methods have to be used to
The summation of the research being conducted in capture and formulate the customer requirements of a
these areas leads to more affordable systems because a design problem. Suitable for those problems are such
comprehensive decision-making strategy has been techniques as ÔSeven M&P ToolsÕ [22] and QFD [8].
utilized. More detail will be provided for these areas in These techniques process the information from the
the following sections. customer requirements directly into an OEC, which
consists of a weighted sum of the aircraftÕs attributes.
Problem Formulation Then a decision as to which is the best design can be
made immediately through an OEC value comparison.
At first, any design methodology has to formulate However, no analysis is involved in this process and
the design problem in a formal, possibly, mathematical only very little information about the design solution
fashion. Hence, information about the design problem can be produced.
needs to be incorporated in a procedure which yields a In doing so, the Òvoice of the customerÓ can be
decision as to the best design. As shown previously in translated to the objectives through such desires as:
Figure 5, the typical design problem has three types of reduction in cycle time, lower cost of ownership,
information: ambiguous, uncertain, and deterministic. dramatic improvements in product quality, reduced
While the deterministic information can be treated with overall life-cycle cost, availability, dependability, etc.
the standard engineering models, such as design and In order to identify the disciplines/sciences needed to
analysis codes, there are no tools readily available to measure and predict affordability, one must examine all
handle uncertain and ambiguous information in design. of the key attributes which contribute to system
Methods of addressing this information include Fuzzy effectiveness.
Logic which is most suitable for the ambiguous
information, and Probability Theory, which most Physics and Process Based Modeling and
suitable for uncertain information [18,19]. The Simulation in VSLCD
methodology introduced in this paper concentrates
predominantly on uncertain information, which is Product (both system and subsystem) models are
captured through random variables. the key to understanding the physical interactions
If the actual value of a design variable is unknown, among various pieces of a complex engineering system.
but there exists some knowledge about the design space, These models often take the form of structural, thermal,
i.e. sample space, it can be modeled as a random fluid flow, or similar physics-based simulation and
variable. This is typically called noise variable [13] analysis capabilities. In contrast, process (both system
and is associated with operational uncertainty. In and subsystem) models focus more on the processes in
addition to the uncertain information about the value of which the product is involved and capture the process

8
American Institute of Aeronautics and Astronautics
impact on various design objectives. Such processes To correct problem accuracy and to reduce
can encompass manufacturing, economics, maintenance, variability, higher order methods can be used for more
etc. Many of todayÕs existing product and process detailed analysis (e.g. computational fluid dynamics).
models were developed and have matured in a single, Though more accurate, these methods require more
focused disciplinary area (e.g. legacy codes). They are problem setup and analysis time with fewer iterations.
often slow, require significant user interaction, and are Current industry practice is to reserve their use for
difficult to incorporate into an integrated synthesis downstream design processes. To mitigate this effect,
process. In other cases, models required for system the proposed methodology includes statistical techniques
synthesis might not exist at all. Furthermore, adequate to construct metamodels of the physics-based product
tools rarely exist that facilitate constraint propagation to tools as a function of the most important variables and
the level of assessment. Thus, one of the fundamental integrating the approximating functions (e.g. response
elements needed in the formation of VSLCD is the surface equations, fuzzy logic, neural networks) into the
development of physics-based models of key responses aircraft synthesis and sizing code. The representation of
and/or constraints as a function of design variables. process models requires heuristic approaches such as
These models must be efficient enough to be integrated expert systems. The resulting tools give a designer the
into a sizing/synthesis program. benefit of using higher fidelity information in earlier
A synthesis and sizing tool is, by definition, a design decision-making. There are at least two additional
multi-disciplinary and can be visualized as a number of benefits gained from using approximations. The design
analysis modules linked via a geometric modeling and space region of interest is continuous, allowing for
mission analysis core as shown in Figure 8. A robust design and simulation techniques to be used.
hierarchical architecture is shown in the figure. In the Second, computational cost is reduced to the evaluation
first level, the geometry and mission core is of algebraic expressions. As a final note, the
supplemented by first level guesses, estimates, and approximations need to be well suited to integration
historical trends during conceptualization (e.g., L/D into an overall synthesis framework or strategy. Key
estimates). The next level consists of departmentalized features of design-oriented models are robustness,
first-order methods of low-fidelity analysis based on a flexibility, repeatability, minimal internally generated
minimum configuration description (e.g., panel numerical ÔnoisesÕ, built-in sensitivity analysis
methods). These analyses possess a high degree of capability, and a capability to be automatically executed
variability in their solutions due to oversimplifications (batch-style or little user interaction).
and failure to capture complex phenomenon currently
only discovered downstream in the design process. Decision Support in VSLCD
These methods are frequently combined into a synthesis
During a life-cycle design process, customer
and sizing tool or executed as off-line analysis and
requirements are transformed into a marketable solution.
implemented as table-lookups.
Moreover, the formulation of the problem will change
Safety as knowledge about a product is acquired and decisions
Aerodynamics
Aerodynamics
Safety

Economics
Economics are made. The evolution of a design is depicted in
Geometry

Synthesis & Sizing

Figure 9. This formulation is multi-level and
Structures
Structures Mission Manufacturing
Manufacturing
hierarchical as complex problems are decomposed.
Integrated Routines
Multidisciplinary and partitioned problems require the
Table Lookup Increasing
S&C Performance
Sophistication and
Complexity
coexistence of multiple decision-making processes that
Conceptual Design Tools
Approximating Functions
(First-Order Methods) are performed simultaneously. This represents the
Direct Coupling of Analyses
S&C
Propulsion
Performance
subsystem problem solution which occurs as product
design moves from conceptual through preliminary and
Preliminary Design Tools
(Higher-Order Methods)
into detail design. Problem management requires an
Propulsion explicit decision-support process. At each product
Figure 8: Varying Fidelity of Synthesis and evolution milestone, any of a number of decisions are
Sizing possible, and the actual selected set of decisions forms a
path.

9
American Institute of Aeronautics and Astronautics
Decision Support Environment Integration in VSLCD
Design decision-making is organized into discrete Engineering simulation via a virtual design
milestones. These milestones are depicted in Figure 10 environment is a key part of the proposed architecture.
as specific steps in product evolution. At each The architecture must sufficiently mask computing
milestone, a decision-making process occurs as technologies as to promote decision-making based on
represented by the decision path in the breakout located the ideas discussed here. Information technology plays a
at the bottom of the figure. Each node of the decision significant role in the preliminary implementation the
path represents a candidate decision to be made by a authors have developed. Key technologies have been
designer. One mechanism for describing a decision is in devised that facilitate the integration and simulation of
terms of a utility function. The utility function the elements of VSLCD and are based on accepted
provides a gauge of product usefulness for a decision. Internet practices where applicable.
Finally, decisions guide product development and, thus, Agents are a key facilitator of VSLCD and are
the allocation of resources to further decrease ambiguity programmatic objects which facilitate the integration,
in a design. These resources are deployed through whether direct or through approximating functions, of
modeling and simulation. product and process based analysis models [23].
One governing metric needed by a designer to Designers benefit from agents due to the repetitive and
assess overall improvements with respect to customer monotonous task of program execution and data
and engineering requirements and methods to include archiving is automated. Models are directly combined
independent sub-system decisions is a multivariate into agents and then linked to the architecture. The
utility function. Since the problem under consideration linking is accomplished via a ÔwrapperÕ which provides
contains uncertainty, the specialized von Neumann- a transparent gateway to computing services such as
Morgenstern expected utility formulation is appropriate communications, name service, and platform support.
for investigation over deterministic utility theory [2]. Earlier discussions highlighting mechanisms for
A precise mathematical problem is proposed in integration of first and higher order physics-based as
order to have the capability to calculate, at an instant in metamodels are depicted in Figure 11 as they relate to
time, the utility and constraint functions which are agents. One or more of these techniques is used to
requisite to the execution of the decision making implement each of the required analysis into the
strategy. At any instant in time, the utility of a design architecture.
can be computed based on the current state with respect
to the side constraints. In turn, the time history of Possible
Alternatives

utility is used to allocate resources in the design Potential Decision

Paths

process. In fact, robust design may be viewed as a

subset of utility.

Decision Trees Final

Solution

Finally, the milestones in the decision path can be Actual

Decisions

organized into a decision tree. Decision trees are useful

Product Evolution Milestones
in the absence of crisp mathematical formulations
during the initial phases of decision making, allows Figure 9: Decision-Support for Product
addressing the various attributes and decision steps in a Evolution
fuzzy, probabilistic, and multi-variant manner. In Current Decision-Tree State Utility Analysis
conceptual design, little is known about a design;
Estimate:
decisions are made based on system level metrics and Z(utility)
Given:
sub-system approximations. It is here that concept X,w,Y,S
selection must be made in the presence of a high degree Satisfy:
Constraints
of ambiguity. The risk is that it may be too expensive
to transition a number of alternatives to preliminary
design but potential payoffs may be lost or costly
redesign incurred if a viable alternative is disregard in Intelligent Decision Support Simulation and Modeling

favor of an unfeasible concept. As decisions are made, a Figure 10 : Decision-Making Milestone and
design progresses into later stages and similar Its Solution
arguments are encountered at product sub-system and
component levels.

10
American Institute of Aeronautics and Astronautics
Figure 12 : Feasibility and Viability: The
Need to ÒShift the CurvesÓ
space is shown in Figure 13. This probabilistic aforementioned aspects of design at the conceptual level.
methodology identifies feasible and viable design The premise behind robust design is that the best way
alternatives and proposes the introduction of new to achieve customer satisfaction is to deliver a product
technologies to increase feasibility, if needed. that performs well not only in the environment for
Step 1. Define the Problem to be Tackled Identify which it was designed, but in all environments. Design
objectives, constraints, design variables (and associated for robustness is achieved by finding settings for control
side constraints), analyses, uncertainty models, and parameters which will not only maximize mean
metrics for each discipline and for the system level. performance in some sense, but also minimize the
This involves translating the customer requirements to objective function variance and satisfy all constraints.
the items listed. This is accomplished in RDS by incorporating all
Step 2. Determine System Feasibility At this stage elements essential to the success of the design into an
of the design process, an optimum is not desired. overall framework, with the ultimate goal of
Instead, an estimate of the percentage of the design affordability which is insensitive to changes in external
space which contains feasible alternatives is important. noise factors.
If there is minimal or little chance of obtaining a Under the RDS, an initial statement of robust
feasible design, there is no use in searching for optimal design optimality is as follows. Note that since the
or robust solutions. noise parameters are typically described in terms of
Step 3. Investigate Active Constraints If the probability distributions, it is intuitively obvious that
system achieves an acceptable P(feas), then proceed to the output from this mathematical model must also be a
Step 5. If the system achieves an unacceptably low (or distribution.
zero) P(feas), an investigation must be performed to find
out which constraints are active and most restrictive. m a x i m i z e s = fcn(mean and variance of Z(X,Y)) or
Crisp definitions for the fuzzy modifiers ÒacceptableÓ Prob(Z(Xi, Yj) < zo)
and Òunacceptably lowÓ are at the discretion of the given Z = Overall obj. (measure of merit) = fcn (X,Y)
designer. Xi = vector of i deterministic variables
Yj = vector of j uncertain variables, defined by
Step 4. Infuse New Technologies The infusion of
uncertainty models Wj
new technologies may be required to improve the zo = Target (a particular value of Z) supplied by
P(feas) value. New technologies almost always affect the customer/decision maker
the underlying physics of the design space and not s a t i s f y i n g imposed constraints, design space ranges
necessarily the geometry of the space itself, as defined
through the design variable ranges. These effects may
be beneficial with regards to one metric while 2
Determine System Feasibility 1
detrimental to another. For example, increased use of Problem Definition
Design Space Model

Constraint Fault Tree Identify objectives, constraints,

x
composites might reduce weight while at the same P(feas)
1 design variables (and associated
side constraints), analyses,
time increasing the vehicleÕs cost of manufacture. FPI(AIS) or Monte Carlo x
2
uncertainty models, and metrics
AND
Step 5. Robust Design Simulation (RDS) [1] x
3
...

Steps 1-4 above are concerned with feasibility, since C1 C2 C3 C4

only constraints are considered. When a large enough 3

Examine Feasible Constraint
Cumulative Distribution
feasible space is found, the space can be searched for Y
Space
Functions (CDFs)
P(feas) P
robust solutions. RDS is a systematic procedure for < esmall
Design Space Model

x
finding settings of design variables which maximize 1
FPI(AMV)
or P
C1
N x
the probability of meeting or surpassing a target for the 2 Monte Carlo
C2
P
objective, while satisfying the constraints. 5 x Relax
Robust Design Sim.
3
Constraints? Y
...

C3
¥ Incorporate Uncertainty Models
¥ Examine Viable Space for
Robust Design Using Probabilistic Techniques Robust Solution through RDS

Robust Design Simulation (RDS) is the part of 4

Infuse New Technologies
Relax Active Y
Re-execute Step 3 Constraints
VSLCD (Figure 7) where system level analysis takes ¥ Identify Technologies to Obtain New CDFs ?
¥ Through DoE/RSM and metrics from P N
place, while accounting for uncertainty, business Step 1, obtain RSEs for:
Ci = fcn (discipline metric ÒkÓ- factors)
Ci
practices, economics, synthesis and sizing, technology, ¥ Use Òk-factorÓ RSEs to select app-
ropriate technologies Old Tech.
New Tech.
and environmental constraints. Application of RDS ¥ Technology Readiness Distributions

can be found in [ 1, 6, 14, 21, 25]. A principle

advantage of this construction is that it gives the Figure 13 : Investigating Feasibility and
designer the ability to concurrently consider the Viability of Multidisciplinary Systems

12
American Institute of Aeronautics and Astronautics
 A flowchart for how one would setup and solve this Concluding Remarks
problem in RDS is illustrated in Figure 14.
Traditionally, design is comprised of a simulation code Present day design practices rely heavily on the
(sizing/synthesis or economic analysis) and an incremental changes and improvements of existing
optimization routine which varies the design parameters designs. This approach has been quite successful
to yield an ÒoptimumÓ solution subject to all imposed because risk, controversy, and negative impact have all
environmental and design constraints. On the other been mitigated through iteration. However, achieving
hand, RDS uses the synthesis tool along with significant design advances, as in the case of new
constraints to perform a probabilistic analysis that innovative, out-of-the-box thinking concepts, requires
yields a figure of merit as a measure of robustness (e.g. innovative design and technological improvements.
objective function response (R) mean and variance [26]). These advances come at the price of increased risk and
This figure of merit is associated with a probability uncertainty.
distribution rather than a single point design solution as A paradigm shift from performance-based analysis
is the case with traditional methods [1]. to design for affordability is required to bridge the gap
One of the major obstacles in applying from evolutionary design to revolutionary systems.
probabilistic methods is to accommodate the large Tremendous payoffs can be achieved if this shift occurs.
variety of existing deterministic computer codes used in Cycle-time reduction, minimal variance designs, robust
modern systems design. A generic methodology is solutions, and affordable systems are a few of the
proposed, which utilizes a ÔwrapperÕ that, when linked perceived benefits. A multi-disciplinary, life-cycle
to the selected analyses codes, drives the program and emphasis must be considered if this shift is to occur.
yields the desired results. Based on this formulation, This includes a mechanism for ascertaining design and
probability functions can be assigned to each of those operational uncertainty, including requirement
input variables which are considered to be uncertain and ambiguity, analytical tool fidelity, decision making in
a cumulative probability distribution function for each the presence of conflict and risk, and the ability to
of the desired objectives may subsequently be obtained. forecast and assess impact and readiness of new
Most probabilistic analyses, e.g. MC Simulation [27], technologies.
estimate their probability distribution functions based The authors have proposed a formulation for
on a large number of samples generated over the design Virtual Stochastic Life Cycle Design that has piecewise
space, defined by the random variable ranges. been successful in making the paradigm shift. VSLCD
The use of computer tools allows for an easy provides the ability to infuse new, ÒbreakthroughÓ
perturbation of input values. However, computation technologies into the design process and evaluate their
time to achieve a probabilistic result increases impact in terms of benefit, cost, and risk even before
significantly as design complexity increases. Three the time and expense of developing and maturing this
methods that incorporate such complex computer technology is complete. Information modeling provides
programs in a probabilistic systems design approach the foundation for representing uncertain, ambiguous,
have been described by Fox[7] and are shown in Table I. and deterministic variables are represented. VSLCD
The use of metamodels has found the widest application includes physics-based modeling and simulation to
and has also been used in the past [1, 6,14,21,25,26,28, allow high-fidelity accuracy to be combined with sizing
29]. The use of statistical regression models, based on and synthesis tools. This creates a multi-disciplinary
Taylor series expansions, along with experimental environment with minimal impact on design time.
designs is very popular [1,6,14,25,26,28,29,30,31,32]. Agent technologies provide integration models that
Robust Design Simulation
Subject to
make the modeling possible. Finally, design decisions
Design/Environmental Robust Solutions
are supported by utility functions as a top-level metric
Constraints
for assessing design progress and subsequent resource
Baseline Metrics
m(R)=f(Xi)
allocation. Robust design is a necessary component of
Model
s(R)=f(Xi)
Control
Model
Building Synthesis & Operational
Monte
Carlo
the utility function, and is calculated using probabilistic
R=f(Xi, Zi) Sizing Environment Simulation
Variables
(Xi)
Sample Objectives:
Min. OEC= s(R)¥ m(R)2
methods. These techniques are combined with a five
Max. OEC= s(R)/ m(R)2
Target for OEC:
step feasibility/viability process to enable the
m(Zi) Uncertainties fZi (PDFs) a¥(m(R)-T) 2+(1-a)¥ s(R)
determination of affordable systems.
(Zi)

Figure 14 : Implementing the Robust

Design Simulation (RDS)

13
American Institute of Aeronautics and Astronautics
 Table I: Code Integration for Probabilistic Analysis
Exact
Probabilistic Analysis Through Direct Coupling
Simple
Analysis tool is used in for each calculation of a Monte Carlo Analysis Tool

Simulation
Advantages: Direct analysis results Inefficient
Objective
Input Probabilistic
Disadvantages: Time consuming analysis Variables Method to
Obtain CDF
CDF

Exact

Metamodel Based Probabilistic Analysis Exact

Complex
Metamodel is used to approximate design code and linked to Analysis Tool

Monte Carlo Simulation Approximation

Advantages: Reduced simulation time, straightforward Metamodel
integration
Disadvantages: Metamodel may be statistically inaccurate, Inefficient
Input Probabilistic Objective
limitation in number of variables Variables Method to
Obtain CDF
CDF

Exact

Exact
Metamodel Based Probabilistic Analysis
Complex
Actual design code is linked to approximate metamodel of Analysis Tool

Monte Carlo Simulation

Advantages: Reduced simulation time, direct analysis results Input
Efficient
Probabilistic Objective
CDF
Disadvantages: Distributions are approximated Variables Method to
Obtain CDF
Approximation

Future efforts are focused on providing expanded 4. Southwest Research Institute, FPI UserÕs and
work on utility as a governing function, and its Theoretical Manual, San Antonio, TX, 1995.
associated probability developments and design guidance 5. Wu, Y.-T., Burnside, O.H., Dominguez, J.,
through simulation and soft computing technologies. ÒEfficient Probabilistic Fracture Mechanics
AnalysisÓ, 4th International Conference on
Acknowledgments Numerical Methods in Fracture Mechanics,
Pineridge Press, Swansea, U.K., May 1987.
The authors wish to thank NASA Langley- 6. Mavris, D.N., Bandte, O., ÒA Probabilistic
Systems Analysis Branch (NAG-1-1793), NASA Ames- Approach to Multivariate Constrained Robust
Systems Analysis Branch (NAG-2-1047), and the Office Design SimulationÓ, SAE Technical Paper 97-
of Naval Research (ONR) for continuing support of this 5508.
research. The authorÕs would also like to graciously 7. Fox, E.P., ÒThe Pratt & Whitney Probabilistic
thank Michelle R. Kirby for her contributions to this Design SystemÓ, AIAA 94-1442.
paper. 8. Arnold, Wilbur V., "Designing Quality into
Defense Systems," Defense Systems Management
References College, Fort Belvoir, Virginia, 1994.
9. Fabrycky, W.J., Blanchard, B.S., Life-Cycle Cost
1. Mavris, D.N., Bandte, O., Schrage, D.P., and Economic Analysis, Prentice Hall, Englewood
ÒApplication of Probabilistic Methods for the Cliffs, NJ, 1991.
Determination of an Economically Robust HSCT 10. ÒAIAA Technical Committee on Multidisciplinary
ConfigurationÓ, AIAA 96-4090. Design Optimization (MDO) White Paper on
2. Hazelrigg, G.A., Systems Engineering: An Current State of the ArtÓ, Jan 15, 1991.
Approach to Information-Based Design, Prentice 11. Technology for Affordability: A Report on the
Hall, Upper Saddle River, NJ, 1996. Activities of The Working Groups to the Industry
3. Wirsching, P.H., Wu, Y.-T., ÒAdvanced Reliability Affordability Steering Group, the National Center
Methods for Structural EvaluationÓ, J. of for Advanced Technologies (NCAT), January 1994.
Engineering for Industry, Vol. 109, Feb. 1987, pp 12. Marx, W.J., Mavris, D.N., Schrage, D.P., ÒEffects
19-23. of Alternative Wing Structural Concepts on High

14
American Institute of Aeronautics and Astronautics
 Speed Civil Transport Life Cycle CostsÓ, AIAA Parametric Analysts Conference, Cannes, France,
96-1381. June 1996.
13. Dubois, D., Prade, H., Possibility Theory: An 27. Kleijnen, J.P.C., Statistical Techniques in
Approach to Computerized Processing of Simulation, Part I, Marcel Dekker, Inc., New
Uncertainty, Plenum Press, New York, 1988. York, 1974.
14. Mavris, D.N., Bandte, O., ÒComparison of Two 28. Mavris, D.N., DeLaurentis, D.A., Schrage, D.P.,
Probabilistic Techniques for the Assessment of ÒAn IPPD Approach to the Preliminary Design
Economic UncertaintyÓ, 19th Internation Society of Optimization of an HSCT using Design of
Parametric Analysts Conference, New Orleans, LA, ExperimentsÓ, 20th ICAS Congress, Sorrento,
May, 1997. Italy, Sept 1996.
15. Taguchi, G., Introduction to Quality Engineering: 29. DeLaurentis, D.A., Calise, A., Schrage, D.P.,
Designing Quality into Products and Processes, Mavris, D.N., ÒReduced Order Guidance Methods
Asian Productivity Organization, available from and Probabilistic Techniques in Addressing Mission
American Supplier Institute, Dearborn, MI, 1986. UncertaintyÓ, AIAA 96-4174.
16. Hwang, C.-R., Yoon, K., Lecture Notes in 30. Chen, W., et al., ÒIntegration of Response Surface
Economics and Mathematical Systems, Springer- Method with the Compromise Decision Support
Problem in Developing a General Robust Design
Verlag Berlin Heidelberg, New York, 1981.
ProcedureÓ, Advances in Design Automation, Vol
17. Murman, E.M., ÒConquering the Barriers to
Multidisciplinary EngineeringÓ, MIT, Dept. of 82-2, ASME, New York, 1995.
Aeronautics and Astronautics, 1993. 31. Kaufman, M., et al., ÒVariable-Complexity
18. Ross, T. J., Fuzzy Logic with Engineering Response Surface Approximations for Wing
Structural Weight in HSCT DesignÓ, AIAA 96-
Applications, McGraw-Hill, Inc., New York, 1995.
0089.
19. Ross, S. M., Introduction to Probability Models,
32. Unal, R., Stanely, D.O., Joyner, C.R., ÒParameter
Academic Press, Inc., Boston, 1993. Model Building and Design Optimization using
20. Neter, J., Wassermann, W., Kutner, M. H., Response Surface MethodsÓ, J. of Parametrics,
Applied Linear Statistical Models, Irwin, Washington, D.C., May 1994.
Homewood, 1990.
21. Mavris, D.N., Mantis, G.C., Kirby, M.R.,
ÒDemonstration of a Probabilistic Technique for the
Determination of Aircraft Economic ViabilityÓ,
SAE Technical Paper 97-5585.
22. Brassard, M., The Memory Jogger Plus + ,
GOAL/QPC, Methuen, MA, 1989.
23. Hale, M.A., Craig, J.I., ÒUse of Agents to
Implement an Integrated Computing EnvironmentÓ,
AIAA 95-1001.
24. Hale, M.A., Craig, J.I., Mistree, F., Schrage,
D.P., ÒDREAMS & IMAGE: A Model and
Computer Implementation for Concurrent, Life-
Cycle Design of Complex SystemsÓ, Concurent
Engineering: Research and Applications, Vol. 4,
no. 2, pp 171-186, June 1996.
25. Mavris, D.N., Bandte, O., and Schrage, D.P.,
ÒEconomic Uncertainty Assessment of an HSCT
using a Combined Design of Experiments/Monte
Carlo Simulation ApproachÓ, 17th International
Society of Parametric Analysts Conference, San
Diego, May 1995.
26. Mavris, D.N., Bandte, O., ÒEffect of Mission
Requirements on the Economic Robustness of an
HSCT ConceptÓ, 18th International Society of

15
American Institute of Aeronautics and Astronautics

View publication stats