AIAA has established a Technical Committee for Multidisciplinary Design Optimization (TC-MDO) with the following charter:
"To provide an AIAA Forum for those active in development, application, and teaching of a formal design methodology based on the integration of disciplinary analyses and sensitivity analyses, optimization, and artificial intelligence, applicable at all stages of the multidisciplinary design of aerospace systems".
One of the functions the TC-MDO established for itself is to provide the aerospace community with a periodic assessment of the state-of-the-art in its field beginning with this White Paper.
The task of developing this initial White Paper was led by Daniel Schrage assisted by Todd Beltracchi, Laszlo Berke, Alan Dodd, Larry Niedling, and Jaroslaw Sobieski.
All members of the TC/MDO reviewed several drafts of the White Paper in its editorial process. A list of the TC-MDO members is included as Appendix II.
This White Paper's purpose is threefold. First, it explores the need for bringing the diverse disciplinary design technologies involved in development of aerospace vehicles and expounded upon in the other chapters in this volume into a concerted action. This approach is necessary to create advanced aerospace vehicles that must be competitive not only in terms of performance, but also in terms of manufacturability, serviceability and overall life-cycle cost effectiveness. Second, it reviews some of the recently evolved means by which such concerted action may be implemented in a systematic and mathematically-based manner referred to as the Multidisciplinary Design Optimization (MDO) technology. Third, it points out major directions for research and development.
The discourse is divided into six sections. The first section presents the need for the MDO technology in the historical context of progress in aerospace. In the second section, the emphasis is on the multidisciplinary nature of the aerospace design process. The human element in that process is discussed in the next section as the key component in any design-oriented technology. The fourth section is devoted to computing as the essential part of the design infrastructure. In the fifth section, the attention shifts to sensitivity analysis and optimization methods that form the core of the MDO technology. Finally, the concluding section identifies the development directions for realization of the MDO benefits.
A. History of Aerospace Systems Design
B. The Need for MDO
II Multidisciplinary Aspects of Design
A. Engineering Design Disciplines
B. Concurrent Engineering Disciplines
C. Supporting Disciplines
III Human Interface Aspects of Design
A. Design Decision Making
B. Meta Design
IV Computing Aspects of Design
A. Information Architecture
B. High Performance Computing
V Optimization Aspects of Design
A. System Level Optimization
B. Decomposition and Sensitivity Analysis
C. Concluding Remarks on Optimization
VI Transitioning to the MDO Environment
VIII The Role of the AIAA MDO TC
Appendix I: Survey of the Industry MDO Practices
Appendix II: AIAA TC MDO Membership Roster
During the pioneering years of aviation, the aircraft designer
frequently was the central figure and the jack-of-all-trades --
designer as well as main resource person in aerodynamics, structures,
materials, propulsion, and manufacturing, often also test pilot,
entrepreneur and founder of great enterprises. The Wright Brothers,
Glen L. Martin, Breguet, DeHavilland, Fokker, Heinkel and Sikorsky
are just a few of the names which come readily to mind. Creative
spirit, clear grasp of essentials, and confidence-inspiring, self-assured
personality were their characteristic traits. The knowledge necessary
to design an airplane was of a practical kind and for many years
it was no more than could be stored in the mind of a capable individual.
This first period came to an end in the early 1930s. Evaluation
of wind tunnel tests in aerodynamics, thin shell analysis in structures,
thermodynamic efficiencies in propulsion, processing and forming
techniques in production - each of them developed into a field
of specialization. The design engineer could not possibly keep
abreast of all developments and had difficulty coordinating the
different inputs coming from various specialists. Yet the solid
engineering background and the long experience of the typical
design engineer provided the know-how and the balanced judgment
to translate new theoretical knowledge into flying hardware. Thus
the senior design engineer had to evolve into what would today
be called the systems engineer. This period lasted from the years
of exciting technical progress in the 1930s, through the years
of mass production during World War II, to the expansion of air
transportation in the 1950s. A few prominent names during this
period are Johnson, Northrop, McDonnell, Douglas, and Hughes.
This period of time also produced rocket pioneers, such as Goddard,
Oberth, Korolev and Von Braun.
In the late 1950s a slow change in attitude occurred throughout
aircraft design. Partly due to the impetus given by missiles,
rockets, and spacecraft which are one of a kind single use systems
that used a new set of design guidelines, and partly due to the
demands of the military who were striving for maximum performance,
the importance and prestige of analytical specialists soared.
Specialists were needed to expand the limits of scientific knowledge
and to reach for ever higher performance. The best minds were
attracted by the challenges of research and development which
usually meant estrangement from design. As a result, the design
engineer's prestige declined. The analytical specialist was often
the originator of novel ideas and the design engineer became the
implementor as he translated these ideas into practice.
Then, around 1970, began the big slump in the aircraft industry
coupled with a decline in the civilian and military space programs
which led to a reduction of the engineering force by about 25%.
Simultaneously, two developments of great potential impact and
far-reaching effect on aircraft design began to take place. First.
computer-aided design came of age and has now relieved the design
engineer of much of the earlier drudgery regarding the menial
aspects of design. Second, the procurement policy of the military
underwent a thorough change. The earlier drive of maximum performance
had been superseded by a new quest for balance among performance,
life-cycle cost, reliability, maintainability, vulnerability,
and other "-ilities". This trend is reflected in the
design requirements growth for advanced aeronautical vehicles
in Figure 1. A major reason for this emphasis
was the control of life cycle costs which are determined by the
design concept and thus are very difficult to change significantly
past this stage as illustrated in Figure 2.
The experience of the 1960s had shown that for military aircraft
the cost of the final increment of performance usually is excessive
in terms of other characteristics and that the overall system
must be optimized, not just performance. The same lesson had been
learned earlier by the airlines when meticulous cost accounting
had pointed toward possible savings due to improved reliability
and maintainability [1]. Cost- effectiveness
for an airliner is mostly economic. The aircraft must generate
sufficient revenue in excess of operating costs that the purchase
investment is more profitable than investing the same amount of
money elsewhere. A similar shift of concern toward cost, supportability,
launch availability, and reliability in orbit began to occur for
similar reasons more than a decade earlier in the space launch
vehicles and spacecraft.
The 1980's brought about a number of thrusts both in government
and industry to improve U.S. productivity and the quality of products.
There has been an on-going quiet revolution in industry for the
past ten years to make the necessary corporate, organizational
and technical changes to compete successfully in an increasingly
competitive global marketplace. These changes occurred first in
the automotive and electronics industries, which were receiving
intense competition for their products from Japan, but in the
late 1980's had spread to the Aerospace industry. Many of the
initiatives in government, particularly the Department of Defense
(DoD), can be traced to recommendations from President Reagan's
Blue Ribbon Commission on Defense Management (Packard Commission)
for improving the weapon system acquisition process. Policy formulation
from these recommendations has come in the form of general acquisition
streamlining and the Total Quality Management (TQM) Program. Other
initiatives can be traced to the DoD's desire to take advantage
of emerging information and computing technologies and the environment
they provide. The DoD - initiated Computer-Aided Acquisition and
Logistics Support (CALS) Program is one example.
As these initiatives have been implemented, there has been increased
realization that in engineering, especially design, lies the greatest
opportunity to improve product quality and provide concurrency
of product and process phases to reduce development time. This
realization has resulted in the recent emphasis on concurrent
engineering (CE). CE has been defined as a systematic approach
to the integrated, concurrent design of products and related processes,
including manufacturing and supportability [2].
This definition is intended to emphasize from the outset consideration
of all elements of the product life cycle from concept through
disposal, including quality, cost, and schedule with traceability
to user requirements. In most cases CE is envisioned as a modem
application of systems engineering in an integrated computing
environment. To date the CE emphasis has been on concurrent consideration
of the life cycle phases, as illustrated in the top half of Figure 3,
for the two-fold goal of improving quality by allowing the natural
coupling among these phases influence the design decisions, and
compressing the overall design process timetable.
Close examination of the Design Phase of the CE process reveals potential benefits from rearranging the traditional disciplinary tasks from the conventional sequential order into concurrent activities shown in the bottom half of Figure 3. The designer can exploit the synergism of the interdisciplinary couplings provided that effective mathematical tools and methodologies are available. Thus, the Multidisciplinary Design Optimization (MDO) methodology that combines analyses and optimizations in the individual disciplines with those of the entire system is a technology that enables extension of the CE concept to the Design Phase.
Design consists of a hierarchical sequence of steps. It begins
with ideas, missions and concepts, takes successively firmer shape
until the configuration can be frozen, continues with the practical
considerations about hardware, and leads to a set of manufacturing
instructions and airworthiness documentation. This evolutionary
process usually is depicted as phases from conceptual to preliminary
to detail design and then manufacturing and production, as illustrated
in Figure 4. As this process evolves design
freedom decays rapidly while knowledge about the object of design
is increasing as illustrated in Figure 5.
As the design process goes forward designers gain knowledge but
lose freedom to act on that knowledge. It was demonstrated mathematically
in [4] that this natural evolution may lead
to suboptimal designs.
Traditionally, for aircraft and most other aerospace systems,
design synthesis and optimization of the overall conceptual system
has been based on achieving a fuel balance and a minimum weight
configuration through parametric variation of a few critical design
parameters i.e. wing loading, aspect ratio, etc. This aerospace
approach to design synthesis is illustrated in Figure 6.
Since aerodynamics and propulsion are the critical disciplines
to achieving a fuel balance and vehicle performance, they are
emphasized and the greatest level of effort is expended in these
areas as illustrated in Figure 5. As the
system design moves into the preliminary design phase and the
initial configuration is frozen, hardware design considerations
begin to dominate and the structures discipline begins to play
a more dominant role. In the detailed design phase the controls
discipline plays an increasing role as flight dynamics and handling
quality improvements usually are necessary to achieve an acceptable
flightworthy system. Also, the transition to production places
a much bigger emphasis on manufacturing, cost, and to some extent
supportability. The obvious problem with this traditional approach
is the short conceptual design phase with an unequal distribution
of disciplines which does not allow use of design freedom to improve
quality and integrate disciplines for optimization. Also, the
balanced design sought by the requirements growth in Figure 1
cannot be achieved. This was also a major conclusion from a recent
industry survey conducted by the MDO technical committee. The
results of this survey have been included as Appendix I.
In recent years there has been an increased emphasis on integrating
the structures and controls disciplines into the design at an
earlier time. For the structures discipline the increased use
of advanced materials with their flexibility and reliability based
structural design philosophies has been one force for this emphasis.
Another force is the use of composite materials for aeroelastic
tailoring, as it couples a structural detail (using skin fiber
orientation angle) with the flexible wing aerodynamics and, ultimately,
the aircraft performance. The controls discipline has really become
an upfront partner. Control configured vehicles offer significant
opportunities for expanded flight envelopes and enhanced performance
through relaxation of inherent stability margins. Flight control
state of the art is perhaps best epitomized by the space shuttle
digital fly-by-wire control system which provides control of the
vehicle from on-orbit maneuvering, through atmospheric entry,
from Mach 25 to a horizontal landing using blended reaction and
aerodynamic controls. Full authority digital fly-by-wire flight
control has been incorporated in operational military aircraft
such as the F/A-18. Application to civil aircraft, prompted by
potential performance advantages in aerodynamics, structures,
and operations has been initiated. However, concerns over reliability,
maintainability, cost, and integrity of such systems has delayed
its application in the U.S. although the A-320 AirBus has a digital
fly by wire system for use throughout normal flight. Control configured
vehicles offer significant opportunities for expanded flight envelopes
and enhanced performance though relation of inherent stability
margins. In addition, ultra-light-weight actively controlled space
structures offer a weight reduction over conventional space structures.
The ultimate goal of control integration is to maximize total
aircraft performance. This goal can only be achieved by a balanced
multidisciplinary design as portrayed in Figure 7
[5].
Aerospace vehicles are engineering systems whose performance depends
on interaction of many disciplines and parts and whose behavior
is governed by a very large set of coupled equations. In practice,
engineers deal with these equations by partitioning them into
subsets corresponding to the major disciplines, such as aerodynamics,
structures, flight controls, etc. In this process of pragmatic
partitioning, the couplings among the subsets tend to be reduced
in number because it is burdensome to account strictly for them
all. Couplings are retained or neglected judgmentally on the basis
of what is known or assumed about their strength in a particular
vehicle category. Generally speaking, the more advanced the vehicle,
the more such couplings should be accounted for.
Rotary wing aircraft or rotorcrafts are an excellent example of
a highly coupled aerospace system. The multidisciplinary complexity
of a rotorcraft, such as a helicopter is illustrated in Figure 8.
Unsteady aerodynamics and vortex interaction cause excitation
of complex structural dynamics to form a unique aeroelastic phenomenon
which is further complicated by a direct coupling with the flight
control system to trim the aircraft. The interaction that takes
place among the disciplines of aerodynamics, aeroelasticity, structures
and materials, and flight mechanics and controls in a typical
flight condition is a series of feedback loops as illustrated
schematically in Figure 9. The coupling
of these disciplines is illustrated in matrix form in Figure 10
by referring back to the feedback loops of Figure 9.
Principal and supporting disciplines are identified for each loop.
If this off-diagonal coupling was not present, a linear superposition
of research conducted by individual researchers at different locations
could be combined. However, the coupling is strong, requires an
interdisciplinary approach, and is one reason why progress in
advancing rotary wing aircraft technology has been difficult.
A similar coupling problem is evident on other advanced aerospace
systems, although the interaction of disciplines would be different,
such as the aerodynamics - propulsion - structures - controls
coupling in hypersonic vehicles. The design synthesis flow chart
using fuel balance for the Aerospace Plane is illustrated in Figure 11
[6].
While multidisciplinary integration can be associated with the
traditional aerospace disciplines aerodynamics, propulsion, structures,
and controls there are also the life cycle areas of manufacturability,
supportability, and cost which require integration. After all,
it is the balanced design with equal or weighted treatment of
performance, cost, manufacturability and supportability which
has to be the ultimate goal of multidisciplinary integration.
Therefore, the multidisciplinary integration aspects of aerospace
system design include the traditional disciplines of aerodynamics,
propulsion, structures, and controls, as well as the life cycle
disciplines of manufacturability, supportability and cost. The
goal of this total multidisciplinary integration is illustrated
in Figure 12. The changes in Figure 12
from Figure 5 are that the conceptual
designer's time has been doubled to capture more knowledge and
use more design freedom; the detail design time has been reduced
by one third based on the use of more upfront design, and a more
evenly distributed effort of disciplines is provided in the conceptual
and preliminary design phase. The dashed line projection from
the "Knowledge about Design" curve reflects the requirement
that more knowledge will have to be brought forward to the conceptual
and preliminary design phases. The dashed line projection from
the "Design Freedom" curve reflects the need to retain
more design freedom later into the process in order to act on
the new knowledge gained by analysis, experimentation, and human
reasoning. The change in the shapes of the two curves would alleviate
the paradox that was discussed in conjunction with Figure 5.
That change might be achieved through better integration of multi
-and interdisciplinary design, analysis, and optimization. Obviously,
another goal is to reduce the design time in order either to shorten
the process duration or to develop a broader selection of optimized
alternative designs in the constant elapsed time.
A clearly defined objective and sufficient budget to accomplish
it is also required for multidisciplinary integration to work.
The space station is an example of a system where much upfront
design has been performed, but no flight hardware has been built
as the funding has been in a continuous state of flux leading
to one costly redesign after another.
Of course, an aerospace vehicle constitutes an integrated system
by virtue of its physics, thus integration is a physical fact
and hardly needs any advocacy for its existence. Therefore, when
we postulate integration, we advocate research and development
of means to help engineers master the interdisciplinary couplings
and to enable them to exploit the associated synergism, toward
improved efficiency and effectiveness of the design process and
better quality of the final product.
Consistent with the above, an integrated design process may be defined as one in which:
(1) Any new information originated anywhere (in any discipline) in the design organization is communicated promptly to all recipients to whom it matters:
(2) When a change of any design variable is proposed, the effects
of that change on the system as a whole, on its parts, and on
all the disciplines are evaluated expeditiously and used to guide
the system synthesis.
It is evident that (1) relies on the technologies for data management
and graphic visualization, while (2) is based on synthesis, analysis
and sensitivity analysis. Together, the above attributes form
a capability for design optimization to be executed in a symbiosis
of the human mind and the computer.
Since the technologies of (1) are well cared for by other AIAA TC's and thrive on the marketplace, it is logical for AIAA TC-MDO to focus its efforts on the technologies underlying (2) which are much less known and, therefore, underutilized: design synthesis, sensitivity analysis, optimization methods, melding the human mind and computer capabilities, and effective organization of engineering to exploit these technologies.
The traditional engineering disciplines for aerospace vehicles include aerodynamics, propulsion, structures and controls. While these individual disciplines are considered fairly mature for many aircraft applications, there are advances in each discipline, due to theoretical, computational and methodology breakthroughs, that foster substantial payoffs and additional research. Emphasis in recent years, however, has been on the advances that can be achieved with research of the interaction between two or more of the disciplines. Also, new disciplines, such as electromagnetics, for low observability, without a statistical database need to be addressed. For advanced and particularly complex aerospace vehicles this interdisciplinary approach is often essential owing to the strong couplings among the disciplines and subsystems and, again, the lack of statistical data and human experience.
While the engineering design disciplines, their interdisciplinary interaction, and optimization of the product are the primary focus for this technical committee it would be remiss if it didn't address their incorporation in the broader set of Concurrent Engineering (CE) disciplines. As depicted in Figure 5 the addition of manufacturing, supportability and cost to the traditional engineering disciplines constitute the set of CE disciplines, with quality being the CE objective function for optimization. The prerequisite task for that addition is development of realistic, reliable, and easy to use mathematical models for manufacturing, supportability, and cost. In contrast to the traditional engineering disciplines, such models are currently inadequate and this inhibits their incorporation in a formal MDO methodology. Obviously, for military systems cost and operational effectiveness and the tradeoff between them receives high priority [7].
Multidisciplinary design optimization of aerospace vehicles cannot take place without substantial contributions from supporting disciplines. The identified supporting disciplines and methodologies are the Human Interface Aspects of Design, Intelligent and Knowledge-Based Systems, Computing Aspects of Design and Information Integration and Management.
The engineering design process is recognized as a two-sided activity
as illustrated in Figure 13. It has a
qualitative side dominated by the human inventiveness, creativity,
and intuition. The other side is quantitative, concerned with
generating numerical answers to the questions that arise on the
qualitative side. The process goes forward by a continual question-answer
iteration between the two sides. The MDO methodology discards
the "push button design" idea in favor of a realistic
approach that recognizes the role of human mind as the leading
force in the design process and the role of mathematics and computers
as indispensable tools. It is clearly recognized that while conceiving
different design concepts is a function of human mind, the evaluation
and choice among competing, discretely different concepts, e.g.,
classical configuration vs. a forward swept wing and a canard
configuration, requires that each concept be optimized to reveal
its full potential. This approach is consistent with the creative
characteristics of the human brain and the efficiency, discipline,
and infallible memory of the computer.
The middle ground between the two sides of design is occupied by the quasi-intelligent and knowledge-based systems. The area of intelligent and knowledge - based systems deals with a broad variety of ways in which the science and technology of Artificial Intelligence (AI) could contribute to the theory and practice of engineering design. The potential contributions cover much more than what are commonly inferred to as expert systems. Expert systems as generally implemented with current techniques. have very limited means of knowledge representation and deduction. The problems of design synthesis using multidisciplinary design optimization will usually require more powerful abstractions than provided by the current paradigm of expert systems [8].
The engineering process can be viewed as a series of decisions which gradually define a new product in more and more detail. As the product evolves from conceptual to preliminary design, to detail design, and then production, the details of the decision making process change radically but its general nature remains the same. Therefore, it can be seen that decision making is at the heart of design. Many different types of decisions must be made in even the simplest case. One must decide where first to look for similar solved problems, how much time should be spent looking at modifications to past or current designs versus new development, which aspects of the design are most important, and how other disciplines are affected. A schematic of how decision makers, using human expertise and expert systems drive the design process is illustrated in Figure 14. These decisions are made in the design process in an environment of uncertainty and risk. Uncertainties come in various forms and the design team faces both upstream and downstream uncertainties. Upstream uncertainties include, for example: uncertainty in the specification of design requirements. This uncertainty relates to the possibility of modification of the original specification that is being designed to. Such changes occur frequently in weapon systems procurements and cause havoc in the design process in terms of schedule slippage and cost increase. Design of space launch vehicles is fraught with uncertainties as to the future mission parameters that may vary in a broad range or vehicle modifications that result in a stretched design. Oftentimes, downstream uncertainties may reflect a lack of knowledge as to the environment in which the product will be used or uncertainties in future availability of spare parts. Uncertainties in manufacturing processes, such as process variability, are also examples of downstream uncertainties from a design standpoint [9].
Design viewed as decision-making implies the need to plan the
decision-making process. Meta-Design "The design of the design
process" addresses the planning activity. As illustrated
in Figure 6the aerospace industry has
developed a general synthesis and analysis which has proved successful
for developing aerospace vehicles from helicopters to spacecraft.
However, the existing design process has been geared principally
to producing designs optimized for performance considerations
without equal regard to cost, schedule, producibility, supportability
or quality. As illustrated in Figure 12
design decisions and tradeoffs may have to be reordered among
multidisciplines and different decisions may be required. A more
flexible design process than illustrated in Figure 6
is required. Plans for integrating CAD/CAE/CAM tools, analysis
tools, and design data bases should be directed toward executing
a specific concurrent engineering design methodology. The type
of design methodology used will depend on the type of design problem
being addressed. Implementing a different computer integration
scheme for each design methodology would pose a considerable burden
in terms of software development. An alternative approach entails
developing a flexible design system capable of supporting the
activities of methodology development (meta-design) and methodology
execution (design) for multiple design problems. Such a system
would be compatible with the evolving idea of a flexible acquisition
process and would be analogous to a flexible manufacturing system
in that it could be rapidly reconfigured to support products of
many different designs. An analytical approach to meta-design
that involves providing a framework that allows the design methodologies
to be developed and evaluated is addressed in [10].
Computer technologies have been changing the environment of engineering
design. Therefore, these technologies are a major supporting discipline
for MDO. Powerful analysis and simulation programs and CAD workstations
are contributing to better solutions. These developments, in turn,
are creating new difficulties. In an environment where most of
the computer activities still involve stand-alone programs, design
engineers often spend 50-80% of their time organizing data and
moving it between applications. Integrated processing with database
system support should eliminate many of these error-prone manual
activities. Data must be shared between disciplines and within
disciplines with all the applicable quality, consistency and integrity
checks.
It should be emphasized that the MDO methodology calls for extending the type of data available to the designer by the new category of the derivative, or trend data that directly answer the "What If?" questions about the entire vehicle system. Examples of such trend data are the derivative of the aircraft range with respect to the wing aspect ratio, incorporating the aerodynamics-structure interaction, or the derivative of the seat-mile operational cost with respect to the take-off gross weight, accounting for the coupling of the structures, aerodynamics, and propulsion. Since the continual concern about the "what if" questions is what a creative design is all about, having a capability to answer such questions expeditiously and comprehensively will constitute a quantum jump in the design process effectiveness and efficiency.
Several parallel efforts have been and are being undertaken to identify an information framework for integrated design. As a result of a NSF workshop [11], a strong recommendation was made for the establishment of a national research program on engineering information management and suggested that the components include:
Engineering Product and Process Description
Engineering Information Dynamics and Data Models
Very High Level Languages and User Interface Engineering
Decision Support Systems
Conclusions from this NSF workshop were that this research will
require the concerned joint efforts of industry, government and
academia and that it will require multidisciplinary teams from
such areas as engineering, computer science, social science and
mathematics.
Another ongoing effort is the work by the Computer-Aided Acquisition and Logistics Support (CALS)/Concurrent Engineering (CE) Mechanical Systems Framework Subtask Group. They have concluded that the information architecture must allow a large multi-disciplinary group to behave as a tightly knit inter disciplinary team, in a concurrent manner in creating product definition information. This architecture includes: concurrent product and process definition, product development team, product life cycle data, and knowledge of customer needs. The architecture may be seen as consisting of an Enterprise Integration framework and an Integrated Information Management System backbone. The Enterprise Integration includes: Product Definition, Process Definition, Configuration Management. Information Exchange, Team Organization, Validation, Metrics, and Enterprise Policy. These elements are peculiar to the enterprise itself. Yet there is an Information Management System that integrates the elements of the enterprise by means of a shared database environment. This includes: Information Modeling, Tool Integration, Information Integrity, Information View, Information Management, Communication, and Resource Definition. The Subtask Group has been assessing the existing environment for Concurrent Engineering from the above stated perspectives. Key topics include:
1) Information architecture,
2) Data exchange standards, such as the Product Data Exchange Specification (PDES),
3) Design - by - Feature,
4) Object - Oriented data management technologies,
5) Storage of (and access to) properties and constraints, material characteristics, and manufacturing methods; and the ability to create (user-specified) multiple views, intelligent libraries, and part, feature, and process information. A first draft of requirements for concurrent engineering information architecture has been completed by the CALS/CE Frameworks Subtask Group [l2].
The term "supercomputer" is commonly used to denote
computing power, but the definition of power in a computer is
highly inexact and depends on many factors including processor
speed, memory size, and so on. Secondly, there is not a clear
lower boundary of supercomputer power. IBM 3090 computers come
in a wide range of configurations, some of the largest of which
are the basis of supercomputer centers at university, government
and industry locations. Finally, technology is changing rapidly
and with it our conceptions of power and capability of various
types of machines. Therefore, the general term, "high performance
computers (HPC)", is a term that includes a variety of architectures.
One class of HPC consists of very large, powerful machines, principally
designed for very large numerical applications, such as those
encountered in science and engineering. Parallel processing assumes
that a problem can be broken into large independent pieces that
can be computed in separate processors. Currently, large mainframe
HPC's such as those offered by Cray, IBM are only modestly parallel,
having as few as two up to as many as eight processors. The trend
is toward more parallel processors on these large systems. Some
experts anticipate as many as 512 processor machines appearing
in the near future. The key problem to date has been to understand
how problems can be set up to take advantage of the potential
speed advantage of larger scale parallel processing [l3].
A NASA Grand Challenge for high performance computing in aerosciences
has been put forth as the integrated multidisciplinary design
of aerospace vehicles and their numerical simulation throughout
a mission profile [l4]. The goal is to demonstrate
the utility of advanced parallel computer systems, including hardware,
software and algorithms, capable of delivering teraflop performance
for the design of a new generation of aerospace vehicles. Such
a demonstration requires separate developments within a number
of disciplines as well as the tight integration of those disciplines.
Figure 15 and 16
provide some indication of the computational complexity and the
present state of the art for two disciplines: aerodynamics and
structural analysis. The underlying assumption is that a single
simulation must be completed in 15 minutes.
Figure 15 shows a range of configuration complexities from an airfoil through a wing to a full aircraft. Figure 16 also shows a range of computational requirements relative to past and present high performance computers. Again, the configuration complexity moves from a simple laminated material through a component to a full aircraft. The computational requirements implied by these figures are severe in their own right. When one thinks of coupling these and other disciplines that are equally computationally demanding through optimization formulation that requires repeated evaluation of these models the "challenge" is truly "grand" [l4]. To meet that challenge, the MDO technologist recognizes that the usable computing speed is a product of the hardware speed and the algorithm speed. In other words, one cannot get very far by using a multiprocessor computer for executing a method that originated [in a] serial computer environment. It follows that to extract full computational potential from a new type of a computer, one needs to invest a development effort in new solution algorithms comparable to the effort that went into the hardware development itself.
Optimization methods have been combined with design synthesis
and parametric analysis and used in the aerospace industry for
the past forty years. The graphically displayed "carpet plot"
is a characteristic of this legacy. In the first two decades the
most commonly used techniques were graphical methods. Graphical
methods were straight forward and easily understood, and had the
obvious advantage of showing at a glance the entire interval of
interest, calling attention to the function peaks, valleys, and
other instructive features. The important limitation of these
methods is that they can paint such a clear picture for only up
to three or four variables in one figure, and require large computer
resources for generating data points for constructing the plots.
For greater number of variables, the combinatorial explosion sets
in that would multiply the figures into volumes, and volumes into
libraries with the attendant loss of the easy comprehension and
interpretation.
During the past two decades much progress has been made in numerical
optimization that offers an alternative to the above. Any design
can be defined by a vector in multidimensional space where each
design variable represents a different dimension. Since we cannot
see in more than three dimensions, the general case is beyond
our power of visualization. Yet the principle is the same as when
we assume only two variables in a base plane and plot above this
plane a curved surface representing the objective function which
depends on the two variables and which is to be optimized. The
objective function may express cost, weight, range, aerodynamic
or propulsive efficiency, return on investment, or any combination
of parameters. It is subject to functional constraints in accordance
with given relationships between variables and parameters and
to upper or lower bounds of variables. The side constraints define
the permissible part of the curved surface where the optimum value
has to be found, e.g. limits due to minimum sheet thickness, maximum
stress, stalling speed, etc.
Thus, in a formal notation, the quantitative side of the design
problem may be formulated as a problem of Nonlinear Mathematical
Programming (NLP):
(1). " find X such that f(X,P) is at minimum
constrained by g(X,P)< 0 and h(X,P)= 0"
where X is a vector of the design variables and Xmin
and Xmax represent variable bounds, P is a vector of
constant parameters, f is an objective function, g is a vector
of inequality constraints, and h is a vector of equality constraints.
Thus, in contrast to the graphical methods, the MDO technology
mathematically traces a path in the design space from the initial
toward improved designs (with respect to some figure of merit)
and does it operating on a large number of variables and functions
simultaneously - a feat beyond the power of human mind. However,
the visibility of the reasons for the design decisions corresponding
to the twists and turns of the search path remain obscured inside
a "black box". Making these reasons visible to the designer
and presenting graphically the salient features of the design
space is a challenge that the MDO technology must recognize and
meet, in order to inspire confidence in the optimization results.
Post optimality and parameter sensitivity analysis can provide
much information that can raise the confidence of the designer.
The idea of formulating a design problem in rigorous, mathematical
terms, introduced in [15], had spawned a
vast body of literature, including comprehensive survey papers,
e.g., [l6], [17], [18],
and [38], and has become a key component
in the MDO methodology. Consistent with its origin, the MDO methodology
has thrived to the largest extent in design of light-weight, aerospace
structures, but is spreading to other engineering disciplines
and non-aerospace applications. The MDO-type methods were particularly
successful in space flight for trajectory optimization. Optimization
has been applied to trajectory design problems for the past 25
years. Analytic optimization has been applied to solving two and
three burn orbit transfer problems for mission planning (estimating
payload transfer capabilities). Boosters (Space Shuttle, Titan,
Delta, Atlas) and upper stages (IUS, Centaur, PAM) use some form
of trajectory optimization to design flight profiles to maximize
payload (or reserve fuel) to some orbital conditions. Reentry
problems have also been optimized to obtain maximum cross range
or down range trajectories. Additionally the NASP trajectory will
have to be optimized to obtain maximum payload to orbit i.e. improvements
in the structure or engine efficiency will lead to new trajectories.
These individual improvements must be weighed against total system
performance to orbit (or some other objective, cost, reliability,
or maintainability) to determine if the new system is worth the
development cost. It should be noted that optimization does not
remove the designer from the loop, but it helps conduct trade
studies. The users should be [warned] not to accept solutions
without careful examination, because if constraints are omitted
from the problem they can often be violated by the optimization
which can reduce safety factors and lead to system failure.
Formulation of the design problem for a system life cycle or concurrent
engineering concept can be accomplished as a multi-objective optimization
problem [l9]:
(2). " find X such that F(fi,(X,P))
is at minimum constrained by g(X,P) < 0 and h(X,P) =
0; where Xmin < X < Xmax;"
which differs from the single objective formulation in Equation 1
by recognizing a set of individual objective functions fi,
i = 1--->NF, which often may be contradictory. The functional
relation f( ) may be as general as admitting all fi's
on equal footing and rendering F a vector, or as specific as a
weighted sum of the fi's which reduces F to a scalar.
By specifying f( ), the designer defines the desired balance of
the various objectives fi. The multiobjective formulation
represents a translation of the customer's ranked requirements
and goals, via the engineering theories and models underlying
the design concept, into a mathematical statement of the design
problem [9],
[l 8].
Numerical optimization capabilities lag in comparative
fidelity as characterized by the number of variables describing
a design for optimization and for analysis (simulation). Equations
are solvable routinely in analysis for tens of thousand, cautiously
for hundreds of thousands, and as tour de force for over a million
variables. Optimization variables for Nonlinear Mathematical Programming
algorithms can not go beyond a few hundred to describe a design,
unless there is some special problem structure that can be exploited
then the number can be extended to ten thousand. Optimality Criteria
(OC) methods do not have any limitation on the number of variables
and problems with a million variables have been demonstrated,
but they apply only if certain conditions are satisfied, considerably
limiting classes of problems for which OC methods may be used.
For example they are not applicable in problems whose analysis
combines governing equations of very different physical phenomena
as is typical for multidisciplinary applications such as the aerodynamics-structures-vehicle
performance problem. In contrast, in some applications involving
a single physical phenomenon, the OC techniques may be very effective
even though they yield only a close approximation to a constrained
minimum. The classic example of this is the Fully Stressed Design
(FSD) technique that works well for homogeneous material structures
but becomes questionable for structures with material mixtures
of varying strength to weight ratios.
Post-optimization analysis of optimal design for sensitivity of
the optimal solution to parameters P is often useful for quick
assessment of the impact of changes to the original problem formulation
[20],
[21],
[40] .
For instance, if the P values needed to specify the F,
g, and h functions in Equation 1 or Equation 2
may vary in an uncertainty range, it may be practical to optimize
the design for the most probable P first. Subsequently, a range
of new optimum designs may be approximated by extrapolation in
the neighborhood of the nominal design using the derivatives of
the optimal F and X. For example, consider a launch vehicle trajectory
that has been designed to maximize reserve fuel a given mission.
If the mission parameters (payload weight, target orbit, or launch
vehicle specifications) change significantly then the trajectory
for the vehicle must be reoptimized to find the trajectory that
maximizes the reserve fuel for the new mission parameters. The
optimum sensitivity analysis may also be very useful in multi
objective optimization (Equation 2)
for evaluation of the effect of the weighting factors subjectively
introduced for converting a set fi's to a scalar F.
Parameter sensitivity analysis is influenced by numerical conditioning
of the underlying problem and solution accuracy, therefore careful
implementation is required to obtain good results [41].
Why System-Level, Multidisciplinary Optimization?
That question needs to be posed and answered first because a typical
disciplinary specialist often tends to strive toward improvement
of the objectives and satisfaction of constraints defined in terms
of the variables of his discipline. In doing so he generates side
effects that other disciplines have to absorb, usually to the
detriment of the overall system performance. A classic example
is aerodynamic design of a transport aircraft wing for a high
lift-to-drag ratio by increasing the wing aspect ratio that may
result in a structural weight penalty needed to alleviate flutter.
That weight penalty subtracts from the performance benefit of
the high lift-to-drag ratio and may actually result in a lower
performance comparing to a reduced aspect ratio wing.
To examine the issue in more detail, consider first an approach
to airframe structural sizing that is often used for a long-range,
subsonic transport aircraft. It may be summarized as follows:
1. Develop aerodynamic shape optimal for the cruise aerodynamic
performance (basically, maximizing the L/D).
2. Minimize structural weight under the stress and aeroelastic
constraints, including flutter, taking into account that the structural
deflections affect the aerodynamic loads and vice versa.
3. From the cruise aerodynamic optimal shape subtract the structural
deflections obtained for the optimized structure under that condition
to establish a jig shape. This will assure that the ideal aerodynamic
shape will be attained at least at one point during the cruise
leg of the mission.
Let us now see what would happen, if we used this approach to
a supersonic transport (SST) flying a mission depicted in Figure
17 whose Mach number diagram may look as illustrated in Figure
18. It is a subsonic/supersonic mission and let us suppose that
we used the supersonic stage in the above sizing approach. Since
there is only one jig shape, if we use it up for the supersonic
stage, we will end up having to accept whatever shape the airframe
deforms to under the subsonic stage cruise condition. That shape
may be aerodynamically suboptimal and cause a drag penalty of
deltaD1 relative to the shape aerodynamically optimal for that
condition.
If we refer to the subsonic stage in the sizing procedure, we
just move the drag penalty to the supersonic stage but do not
remove it. To remove or, at least, drastically reduce that drag
penalty we have to recognize that there is a three-way mutual
dependence of the aerodynamic loads-structural sizing-deflected
shape that we, as structures engineers can manipulate to our advantage
by changing the structural stiffness, its magnitude and distribution,
over the airframe. Without invoking the notion of formal optimization
as yet, suppose that by judgment we increase the wing stiffness
in the outboard area to reduce the elastic wing twist that contributes
to the drag penalty under the subsonic stage cruise condition
(the optimal supersonic shape has remained optimal because we
compensated by the jig shape). That may cost a structural weight
penalty of deltaWI which is, in general, bad for the performance.
However, if drag is reduced from deltaDI to deltaD2 < deltaD1,
generally a good influence, the performance analysis can be referred
to evaluate the deltaWI against (deltaDl<deltaD2) as a trade-off.
The trade-off may come out positive or negative depending on the
objective and usually there is a wide choice. A few examples are:
the minimum take-off gross weight (TOGW) for given range, payload,
and mission profile; the maximum payload for a given range, TOGW,
and mission profile.
The above trade-off example is also only one of many. Suppose
that the wing is strength-critical in 2.5g pull-up. Then, we may
wish to allow the outboard wing more twist flexibility so that
it can wash out thus alleviating the wing root bending moment
and reducing the structural weight at the price of increased drag
of the wing elastically deformed during the subsonic cruise.
Many such trade-offs have to be considered simultaneously, and
a complicating factor is that they have to be resolved not only
to end up with a positive net impact on the performance objective(s)
but they also have to be solved without violating the constraints
imposed by each of the participating disciplines, e.g., flutter,
allowable stress, vehicle stability, controllability, etc. It
is clear that the human judgment needs help from the computer
for resolution of such a multitude of trade-offs.
Leaving the above example and returning to the generic discussion,
it may be asserted that the user demand that drives the development
of multidisciplinary analysis and optimization has been intensifying
because:
1. major new aircraft design projects become fewer and farther
apart in time, hence the past experience becomes less available
as a guide in making the design decisions;
2. advanced aircraft tend to be an enormously complex system of
interacting parts and disciplines and its ultimate performance
hinges on the myriads of numerical interplay, some of them very
subtly and beyond the power of human judgment to evaluate precisely.
The ubiquitous challenge of design may be phrased as "How
to decide what to change, and to what extent to change it, when
everything influences everything else".
The integrated design process that was defined at the end of Section IB
is intended to meet the above challenge by creating an environment
on the quantitative side of design (Figure 13)
in which the designer's decision making will be supported with
a comprehensive, and quickly generated, numerical information
presented in an easy-to-interpret format. It is not the purpose
of this paper to systematically survey the state-of-the-art in
the methodology for creation of the above environment or to endorse
a particular approach or technique. Rather, its purpose at this
point is to illustrate emergence of a new methodology for multidisciplinary
design optimization by a few examples of methods whose initial
application experience has been encouraging.
It is generally agreed that the challenge posed by the quantitative
side of an advanced aircraft design as a complex system needs
decomposition that breaks the large, intractable problem into
smaller subproblems while maintaining the couplings among the
subproblems. In the design office, this approach maps well onto
the natural organization of engineers into groups by disciplinary
and task specialization. It preserves and nurtures the advantages
of the division of labor, including the concurrency of operations
- the time-honored principle of industrial management first articulated
by Adam Smith in the classic work "The Wealth of Nations"
nearly 250 years ago [23].
The decomposition approach stems from the realization that the
analysis and sensitivity analysis that generate data optimization
algorithms need may easily account for more than 90% of the total
computational optimization cost. Hence the recent emphasis on
the efficient sensitivity analysis that exploits modularity in
application to complex systems. Numerous decomposition schemes
have been proposed in literature and, undoubtedly, more will be
developed in the future. For the purposes of this discussion it
will suffice to name as two basic examples the methods for a hierarchic
decomposition and a non-hierarchic decomposition.
Hierarchic decomposition. The concept of a hierarchic decomposition
for engineering design was introduced in [24]
using the algorithm from [25] as means for
efficient calculation of the optimum sensitivity derivatives.
Examples of this type of decomposition applied to structures may
be found in [26],
and a demonstration of its usefulness in multidisciplinary optimization
to aircraft configuration was given in [27].
The hierarchic decomposition method exploits a special way in
which the computational and decision making operations may be
arranged in the design process of an engineering system. The arrangement
is illustrated in Figure 19. Each box
represents analysis and optimization of a subset of the entire
system problem. The analysis information flow is topdown from
the "Parent" black-box to the "Daughter" black-box.
For example, a finite element analysis of the entire airframe
may be a Parent that transmits the boundary forces to a Daughter
wing substructure and the natural vibration frequencies and modes
to another Daughter representing aeroelastic behavior. The topdown
flow ends when it reaches the bottom level of the black-box pyramid.
Then, each black box solution is available and the optimizations
begin progressing from the bottom level up.
Inputs received by a Daughter from a Parent are frozen as constant
parameters for the duration of optimization performed inside of
the Daughter black-box. Moving up to the Parent, one transmits
the results of the Daughter optimization augmented with the derivatives
of these results with respect to the parameters that the Parent
has sent to the Daughter. These derivatives enable the Parent
optimization to account by linear extrapolation on the effect
of the Parent design variables on each Daughter constraints.
The procedure continues to the top of the pyramid. The top Parent
represents the system level objectives and constraints and is
controlled by the system level design variables. The effects of
these variables on all the black-boxes in the pyramid below are
accounted for by the optimum sensitivity derivatives transmitted
from below. Since the procedure is based on first derivatives,
it takes a few iterations to converge. Each iteration consists
of the analysis sweep top-down and the optimization sweep bottom-up.
With careful implementation the optimization on successive iterations
becomes more efficient if warm/hot start capabilities are used.
Since the Daughters do not communicate at the same level (no information
transmission among sisters), the individual black box analyses
and optimizations at each level may be performed in parallel.
Non-hierarchic decomposition. The non-hierarchic decomposition
method allows for information multidirectional transmission among
the black-boxes forming a system as depicted in Figure 20
for an example of a flexible, actively controlled wing. A system
like this cannot be arranged into a Parent-Daughter pyramid shown
in Figure 19. Its optimization may be
executed as a single operation for the entire system and is guided
by the system sensitivity measured by the derivatives of the system
behavior (response) variables with respect to the system design
variables.
The derivatives may be computed without finite differencing on
the entire system analysis by a technique that:
1. solves the system at a baseline design point,
2. computes the partial sensitivity derivatives of the output from each black-box with respect to its input from other black-boxes and with respect to the design variables,
3. uses the above partial derivatives as coefficients to form
a set of simultaneous, linear, algebraic equations whose solution
yields the system sensitivity derivatives.
A review of various types of decomposition, including the hierarchic
and non-hierarchic approaches, was provided in [28].
The mathematical concept underlying the non-hierarchic approach
was introduced in [29] and [30].
Its applications in aerospace design were compared to that of
the hierarchic decomposition in [31], and
an example of its industrial use was described in [32].
Common to optimization by both hierarchic and non-hierarchic decomposition
is its reliance on the sensitivity analysis as a generic numerical
method in engineering analysis [33] as well
as a disciplinary method of the type described for structures
in [34] and for aerodynamics in [35].
How to decompose a system. When the system at hand
is new and there is no past experience in guiding its decomposition,
one may benefit from the use of a formal technique that converts
a set of randomly sequenced black-boxes into a set ordered into
a hierarchic, nonhierarchic, or a mixed, hierarchic/non-hierarchic
arrangement. The technique formalism requires that each black-box
be defined as a source and a recipient of information. As a source,
the blackbox sends information through its vertical sides, horizontally,
to the left and to the right. This definition is illustrated in
Figure 21.
Initial random sequencing is presented by a diagonal chain of
modules shown in Figure 22. The execution
sequence is initially assumed to proceed from the upper left corner
to the lower right corner and the modules are positioned randomly
along the diagonal. Each off-diagonal dot marks a data interface
indicating that the output moving along the horizontal line is
directed along the intersecting vertical to the recipient module.
The dots in the upper right triangle mean feeding the data forward
(downstream), by the same token the lower triangle dots mean feedback
(upstream). Each instance of a feedback calls for an iteration
because module A upstream depends on the output from a successor
module B downstream.
By a systematic row and column permutation executed by a computer
program, the random picture of Figure 22
may be transformed into an ordered sequencing shown in Figure 23.
The transformation goal was to eliminate as many feedback instances
as possible. It was not possible to eliminate them all in this
particular case. However, their number was reduced and the remaining
feedback instances have been clustered. That clustering suggests
decomposition shown in Figure 24. It
is a hybrid decomposition, hierarchic with respect to the clusters,
each represented by a box in the pyramid, and non-hierarchic inside
each cluster. Software tools became recently available for generating
this type of decomposition from the initial, unorganized set of
computational modules as described in [36].
The above examples of methods now under development and testing
should not be regarded as the last word but only the beginning
in evolution of a new methodology for quantitative support of
the design process. One common thread of the examples discussed
in the foregoing is the concern about creating an environment
in which the engineer's mind and computer interact drawing on
the best resources of each. This concern is expected to alleviate
misgiving some practicing engineers may have about the formal
design methodology that was offered, on occasion in the past as
an "automated design". That was a misrepresentation
that might have been an underlying cause of the lag of applications
behind the theoretical developments noted in the survey in [16].
The other common thread is the concern about modularity of implementation
necessary to ensure flexibility, open-endedness, and ability to
accommodate a variety of the information sources, including judgmental
estimates, statistics, references, and experiments, in addition
to computer programs. Modularity is also seen as a prerequisite
for exploiting the computer technology progress in multiprocessor
machines and distributed computing. Finally, there is a pervading
concern for making the information exchanged among disciplinary
specialists quantified and precise to provide a basis for the
qualitative discourse these specialists are engaged in.
With these concerns in mind, one may foresee further developments
as encompassing new algorithms for decomposition, disciplinary
and system sensitivity analysis, effective search and optimization
of the design space, and AI-based tools making all this user-friendly.
The central role of the disciplinary and system sensitivity analyses
was apparent in the above method examples. Disciplinary sensitivity
analysis by quasi-analytical approach is now routine only in structures
and immediate emphasis is needed on developing a similar capability
in CFD - the other major consumer of computer resources in aircraft
design. The system optimization will become well-rounded when
all contributing disciplines are liberated as much as possible
and practical from the tedium of finite differencing by augmenting
their analyses with sensitivity algorithms. Progress in the techniques
for search and optimization in the design space is also important
for the overall effectiveness and efficiency of the methodology
as are procedures for tying together that search with analysis,
sensitivity analysis, and approximate analysis, including the
approach of statistically-fitted response surface methods. Improvements
in the search techniques are needed for effective identification
of multiple local minima - a vexing problem that thus far lacks
a rigorous mathematical solution for cases with more than a few
variables. One should also keep in the field of view the optimality
criteria as an alternative to the search of the design space.
Finally, the development should be kept open to accommodate innovations
such as the self-learning neural nets, and genetic algorithms,
to mention but a few examples of the cutting-edge approaches.
As always in methodology development, the ultimate test of usefulness
is in applications. Therefore, a systematic cooperation of the
theoreticians, implementers, and users who apply the tools and
influence the theory and implementation with their observations
and wishes must be an intrinsic part of that development. The
benefits from introduction of the new methodology will be amplified
if that methodology is applied early in the design process where
most of the leverage is available.
The previous sections of this white paper have reviewed different
aspects of MDO. This section will provide some thoughts on how
to evolve to a concurrent engineering (CE) environment and the
role MDO for aerospace systems will play in this transition. The
goal is to achieve the compression of the tasks in the Design
Phase illustrated in Figure 3 and redistribution
of the effort among the engineering disciplines as indicated by
the horizontal bars in Figure 12. The
expected end result is more design freedom retained longer into
the design process and more information about the object of design
gained earlier in the process as portrayed by the curves in Figure 12.
To accomplish the above one needs to develop an environment for
the integrated design process as defined at the end of Section IB
The following specific tasks should constitute that development:
1. Identify information exchange requirements - each discipline
describes its input and output information.
2. Establish unified numerical modeling parameterized in terms
of the design variables - a consistent vehicle geometry must be
the basis for all mathematical models, and changes to the geometry
must be centrally coordinated.
3. Establish a data management system for a quick and easy location
and transfer of the information needed by the engineers and by
the computational tasks, and for generation of good initialization
data for the optimization tasks.
4. Develop mathematical models for manufacturing, reliability,
supportability, and life cycle cost, to augment the classical
discipline models for a complete implementation of the CE idea.
5. Assemble an efficient design-oriented analysis capability.
A design-oriented analysis is tailored to support applications
in design characterized by: repetitive use with only a subset
of the input changed in each repetition, need for sensitivity
data, use of the mathematical models of varied degree of refinement
to trade accuracy for computational cost.
6. Efficiently generate discipline design sensitivities.
7. Assemble a system sensitivity analysis for vehicle optimization
- system design variables will be identified and used to quantify
the effects of design changes on the system behavior.
8. Improve optimization algorithms for effective handling of very
large number of design variables, disjoint and nonconvex design
spaces, multiple minima, and multiobjectives.
9. Improve post-optimum sensitivity analysis for greater computational
efficiency, and for effectiveness in the extrapolations across
the points where the set of active constraints changes its membership
(see [20] for the description of a problem
caused by changes in the active constraint set).
10. Develop a method for systematic developments and evaluation
of design changes toward meeting the objectives and constraints
in form of an iterative, multidisciplinary optimization process.
The above development will result in a new, higher level of the
state-of-the-art in engineering design. It is anticipated that
industries, government laboratories, and universities will all
contribute building blocks. There will be an accumulation of generic
and proprietary, product-tailored tools, and of partial implementations
of the entire process. Pilot projects will accumulate experience,
demonstrate benefits, and build confidence. Gradually, a complete,
new, integrated design process will evolve and be used for creating
aerospace vehicles.
That process will be a logical expansion to the Design Phase of the CE concept defined in [37]. The most important ten CE characteristics from the above reference (slightly rephrased for the context of this discussion) and their relationship to MDO are listed in Table 1 to emphasize once again the view of MDO as a key new component in CE. In that development, the AIAA TC-MDO has a role described in Section VIII.
| 1 | Compreh. Sys. Eng. Proc. Using Top-Down Design Approach | Authoritative, but Particip. Top Mgt ; System Eng. Mgt. Plan (SEMP) ; Automated Config. Mgt/Control | Decomposition |
| 2 | Strong Interface with Customer | Methods for Translation of Voice of Customer Into Prod/Process Characts. | Optim. Methods |
| 3 | Multi-Function Sys. Eng. and Design Teams | Management and Peer Acceptance; Equal or Near Equal Analysis - Cap. | Decomposition and Sensitivity Analysis |
| 4 | Continuity of the Teams | Training org. accept and Incentive Program | |
| 5 | Practical Eng. Optim. of Product & Process Characts. | Methods for Incorp. Qual. & Quant. Optim. Methods | Compat. of Num. Optim. Methods with Other Methods |
| 6 | Design Benchmarking Through Creation of a Dig. Prod. Model | Design by Feature Methods Plus Data Exchange Stands | Sensitivity Analysis and Optim. Methods |
| 7 | Simul. of Product Perf. and Manuf. Process | Destrib. Simul. Cap. with Varying Levels of Fidelity | Sensitivity Analysis and Optim. Methods |
| 8 | Experiments to Confirm/Change High Risk Predictions | Design of Experiments Methods for Variability Reduction of High Risk Characs. | |
| 9 | Early Involvement of Subcontractors/Vendors | Accept. by Top Mgt. and Peers Plus Organ. Decomposition | Decomposition |
| 10 | Corporate Focus on Contin. Improve. & Lessons Learned | Design Tracking and Library Access through an Autom. Config. Mgt./Control System | Decomposition, Sensitivity Analysis and Optim. Methods. |
Multidisciplinary Design Optimization (MDO) has been rapidly gaining
recognition as a new, engineering discipline that assumes a key
role in development of advanced aerospace vehicles whose common
characteristic is that they are complex engineering systems. In
its role of a catalyst and conciliator of the disciplinary requirements
and interactions, MDO becomes as important for success of design
as any traditional engineering disciplines. MDO has been reviewed
in the historical context of the aerospace design process evolution
and in the context of the present day and future challenges posed
by advanced aircraft and spacecraft. If this White Paper were
written a decade ago, in all likelihood it would have emphasized
design optimization for improved performance. The recently evolved
understanding that performance is only a subset of the overall
product quality that must include the cost of development, manufacturing,
and maintenance has replaced that emphasis in this paper with
one that includes the entire life cycle of the aircraft or spacecraft,
with the cost of that life cycle as one of the key objectives.
This meshes very well with idea of Concurrent Engineering whose
main goal is to move the manufacturing and supportability considerations
upstream into the design process in order to compress the entire
development and to assure that these considerations get in the
design process an attention equal to that traditionally afforded
the vehicle performance. This basic idea of Concurrent Engineering
- the compression of the major life cycle phases of Design, Manufacturing,
and Maintenance that were sequentially arrayed heretofore - applies
also to the phase of design. That phase also may be "compressed"
in the sense of staggering the conventional sequence of operations
and decisions.
MDO is seen as a means by which to achieve the above compression
by bringing more information about the entire life cycle and the
vehicle performance and cost aspects earlier into the design process.
This will enable engineers to make design decisions on a rational
basis that gives equal consideration to all the influences disciplines
exert on the system, directly, or indirectly through their complex
interactions. Doing this early in the process exploits the leverage
of the uncommitted design variables. On the other hand, it is
equally important to extend the MDO-based approach to the later
phases of the design process in order to take advantage of the
new information that becomes available during that process through
creative thinking, analysis, experimentation, and exploration
of alternatives. In order to do that, the design variables that
in the conventional design process are decided and set early,
need to be retained as free variables much longer into the process.
Using the MDO technology one may achieve this because the overall
methodology of system analysis, and optimization based on sensitivity
data remains the same throughout the process. The variable element
is analysis that deepens as the process moves on.
The MDO methodology is well-suited to blend in the above analysis
the traditional, performance-oriented design considerations with
those posed by the remainder of the life cycle because it is generic
and capable of including anything represented by a mathematical
model, whether that model is derived rationally or established
heuristically. However, it is necessary to develop such models
first and this is one of the several specific developments identified
in the White Paper. Another development direction of a high pay-off
potential pointed out is toward the probabilistic methods, multiobjective
capability, and facilities to accommodate the "soft"
(negotiable) constraints as distinct from the hard constraints
in optimization - as required by the applications of MDO extended
to manufacturing, maintenance, and economics.
The key premise expounded for the MDO approach in the White Paper
is that it is not a "push button" design. Instead, MDO
is an environment in which the human ingenuity combines with the
power of mathematics and computers in making design decisions.
The boundary between the formal mathematical methods and the human
judgment is, of course, fluid. Nothing should prevent an engineer
either from delegating a repetitive tedious routine to a formal
method or from substituting judgment for a formal method or from
overriding the method results.
Based on that premise, the MDO-enhanced design process has the clear potential for radically improved product quality achieved by systematic exploration of the alternatives created by human ingenuity and bringing each of these alternatives to the optimal state among which a fair choice can be made by engineer's judgment.
The TC-MDO should be a focus for MDO activity, providing a forum
through which the efforts of researchers can be disseminated to
users and potential users in industry and government establishments.
At the same time, feedback from users will establish future requirements
and goals.
In order to maintain such a forum, the TC should seek membership among all engineers and computer specialists, involved in design, and design support, of aerospace vehicles of all major categories such as aircraft, launch vehicles, spacecraft, missiles, transatmospheric vehicles, etc.
To achieve the goals called for by its charter, the TC should
undertake the following tasks:
(1) DEFINE the technological sphere of interest in multidisciplinary
design optimization regarded as a new engineering discipline and
one of the key elements in concurrent engineering and total quality
management.
(2) GATHER information on MDO
- university research
- industry practices and applications
- government research and requirements
(3) EDUCATE
- upper and middle management in industry and government
- R&D engineers in industry and government
- university graduate and post graduate students
(4) GUIDE research efforts by suggesting areas for study, and
future goals.
To accomplish these tasks the following TC-MDO subcommittees have
been formed:
(1) White Paper - Act as a focal point for a periodic generation
of a white paper expressing the collected views of the TC and
describing state-of-the-art in integrated MDO.
(2) Computer Technology and Optimization - Act as a focal
point for information concerning optimization algorithms and their
application and advances in computer technology.
(3) Education - Act as a focal point on all issues relating
to education in MDO.
(4) Liaison - Act as a focal point to coordinate activities
and provide a channel of communication with other active AIAA
TC's.
(5) Conference Support - Act as a control focus of activity
and resources of the TC-MDO in support of AIAA sponsored and co-sponsored
conferences, symposiums, and shows.
(6) Publications - Act as a focal point for generation
and distribution of all publications of the TC- MDO.
(7) Benchmark - Act as a focal point for devising effective
and practical test cases for MDO methods.
(8) Emerging Methods - Act as a focal point for identifying
emerging methods applicable to MDO.
(9) Material Optimization - Act as a focal point for coordinating
research efforts in the area of optimum design of materials, and
their inclusion into the design of complex systems together with
the other relevant disciplines.
(10) Awards - Act as a focal point for identifying and recognizing significant contributors to MDO.
1. U. Haupt, "Decision-Making and Optimization,"
NPS-67 Hp 77021A, February 1977.
10. J. E. Rogan and W. E. Cralley, "Meta
Design," IDA, Paper P-2152, Jan 1990.
17. Siddall, J. N.: "Frontiers of Optimal
Design," ASME J. of Mechanical Design, Oct. 1983.
22. Aviation Week and Space Technology, June 18,
1990, pp. 98.
23. Adam Smith: "The Wealth of Nations,"
1791.
27. Wrenn, G. A.; and Dovi. A. R.:
"Multilevel Decomposition Approach to the Preliminary Sizing
of a Transport Aircraft Wing," AIAA Journal of Aircraft,
Vol 25, No 7, July 1988, pp. 632-638.
In the summer of 1990, the AIAA Technical Committee for Multidisciplinary
Design Optimization conducted an industry survey on the use of
the MDO technology. The survey was taken to their companies in
the U.S.A. and in Europe by the TC members who used their company
contacts to answer the survey questions. Thus the answers received
were representative of the company rather than individual opinions.
The first part of this appendix defines the survey purpose and
background. A Summary of the results is given in the second part.
Survey Definition
Purpose The survey purpose is to determine the ways and
means the aerospace industry uses to resolve trade-offs that arise
in design process of aerospace vehicles, with emphasis on the
trade-offs that involve two or more engineering disciplines.
Background The following examples illustrate the notion
of a trade-off. By increasing the aspect ratio of a transport
wing, the drag-due-to-lift is reduced thus improving range for
a given payload. However, a higher aspect ratio wing, in general,
will weight more tending to decrease range. The net effect of
change in aspect ratio on range may then be positive or negative,
depending on the strength of the drag and weight influences.
The kill probability of an air-to-air missile may be increased
by making the missile more agile, or making the fighter that launches
the missile more agile, or both. There is a cost associated with
adding agility to the missile and another cost of adding agility
to the fighter. In what proportions should one allocate a fixed
total budget to the missile development and to the fighter development
to get a missile/fighter system of the maximal kill probability?
The pointing accuracy of a large antenna dish attached to a spacecraft
constructed as a large, actively-controlled structure, may be
improved by making the structure more rigid, or by adding more
capability to the control system. There are weight penalties,
and cost penalties for both alternatives. What is the "best"
mix of added structural rigidity and added capability of active-control
system to achieve the required pointing accuracy?
As the examples illustrate, the trade-off arise at high-level
(system level) as well as more detailed level, in all classes
of vehicles. For proper resolution they involve numerical information
and judgment.
Regarding numerical information, there is a body of mathematical
methods such as: disciplinary and system analyses, sensitivity
analysis (to compute derivatives of the dependent variables with
respect to independent variables by analytical, quasi-analytical,
or finite difference techniques), parametric studies, and formal
optimization. On the judgment side, the approaches range from
unstructured decision making to highly organized and disciplined
procedures for generation, evaluation, and recording of the judgmental
decisions.
It is not clear, however, where the center of gravity lies between
the extremes of the all mathematical and all judgmental ways of
resolving the trade-offs, and what are the most often used techniques
in both categories. It is also not clear whether things are as
they should be with regard to the above, or whether they should
be changed. It is important to know the industry opinion on this
issue for effective planning and development of the pertinent
methodology and engineering education. This survey should shed
some light on the issue.
Format The survey subject is really too complex to boil
down to a simple, check-a-box, questionnaire. Therefore, a free
format essay is preferred (please, include identification of your
company, your position, and give an example of a product to which
the issues raised in this survey would, typically, apply). The
minimum length for a meaningful answer is probably less than one
single-spaced page. To facilitate the evaluation, the maximum
length should not exceed 3 pages. However, a questionnaire format
is also available, if time for a free-format answer cannot be
found.
Summary of the Survey Results: Questions and Answers.
Most of the survey returns came in the Questionnaire Format but
several were in an all free-format narrative. The survey Questionnaire
Format questions are reproduced in full. Most questions called
for a numerical answer. The numerals following each question represent
averages of the survey return. The answers were also illustrated
by placing the averages on the numerical axis. Since there is
no uniform definition of design stages, the answers were classified
as pertaining to early and late phases of design and marked by
E and L, respectively. The averages include also the information
extracted judgmentally from the free narrative results. Questions
4 and 6 in the Questionnaire called for free-format answers and
are followed by paraphrased extracts from these answers and from
those returns that came in an all-free-format narrative.
1. Assuming a scale from -5 (all mathematical) to +5 (all judgmental),
place on the scale the center of gravity of the ways by which
the design trade-offs are being resolved, for each design stage.
Notes: 1) results are reported for early/late design stages, 2)
"system" means a complete vehicle.
-1.1/ -2.2 (early/late)
Mathematical.......|.........Judgmental
-5...-4...-2...-1...0...1...2...3...4...5
...........L....E....|........................
2. In the judgmental decision making, where is the center of gravity
between the extremes of very formal organizational procedures
(-5) and unstructured process (+5). Use a format as in answer
1.
-1.1/ -2.2
Mathematical.......|.........Judgmental
-5...-4...-2...-1...0...1...2...3...4...5
...........L....E....|........................
3. For the numerically generated information, please, evaluate how much does your organization rely on the following mathematical tools, using a scale from 0 (not used) to +5 (used very often, regarded as essential).
Analysis
Disciplinary analysis 4.2/4.4
0.....1.....2.....3.....4.....5
...........................EL
System sensitivity by parametric study: 3.0/3.5
0.....1.....2.....3.....4.....5
...................E..L.........
System sensitivity by finite differences: 2.8/1.5
0.....1.....2.....3.....4.....5
..........L......E..............
System sensitivity by analytical/semi-analytical method: 3.0/2.3
0.....1.....2.....3.....4.....5
...............L...E............
Optimization
Parametric study/disciplines: 4.0/3.5
0.....1.....2.....3.....4.....5
.......................L.E......
Parametric study/system: 4.2/4.2
0.....1.....2.....3.....4.....5
..........................L/E...
Formal numerical optimization/disciplines: 3.0/2.8
0.....1.....2.....3.....4.....5
..................LE............
Formal numerical optimization/system: 3.0/2.0
0.....1.....2.....3.....4.....5
.............L.....E............
4. If formal, numerical optimization is used, name a few techniques,
e.g., nonlinear programming (NLP), linear programming (LP), optimality
criteria, and names of a few optimization programs (Early/H for
in-house developed, A for acquired from outside).
NLP, LP, Fully Stressed Design, Optimality Criteria (FASTOP),
Design of experiments (DOE), Mix of in-house and acquired, Most
of NLP at early stages, little in Aerodynamics, OC and FSD at
later states in Structures. Formal optimization of the configuration
in early stages, after that structural optimization with the configuration
frozen.
5. For each design stage indicate whether the present system adequately
identifies the best design options and configurations, accounting
for complex interactions among the system parts and governing
disciplines. Use scale from 0 (very inadequate) +5 (completely
adequate).
2.9/3.2
0.....1.....2.....3.....4....5
...................E.L.........
6. Finally, indicate whether you are satisfied with status quo
or would like to see a change.
Formal optimization applied to configuration (system) very early,
then configuration frozen, optimization limited to structures
and control.
The above confirms the paradox: In the design process, "the
knowledge increases with time, the freedom to act on that knowledge
decreases with time".
Present ways adequate to design good vehicles, not adequate "to
prevent problems from occurring late in the design cycle which
require costly and sometimes futile efforts to correct".
After the configuration is frozen, problems arising in a particular
discipline are expected to be solved by a fix limited to that
discipline (e.g., flutter fixed by stiffening of the wing structure
or by balance masses).
Organizational structure and culture must change to bring about
an effective MDO into the design process.
The best place for MDO is in the middle of the design process
when enough hard information is available but before too many
variables get frozen and before the problem size mushrooms.
Better infrastructure is essential: faster, bigger computers,
visualization, data bases.
Lack of the system sensitivity information hampers the design
process.
"Higher order" disciplines (e.g., aeroelasticity) are
particularly limited by the above.
High priority should go to a complete automation of the routine
engineering tasks, including AI methods.
MDO should be used at ALL stages of design
Mathematical models of different degree of refinement should be
used in a coordinated manner throughout the design process.
Doing work faster = the MDO advantage.
Need a better handle on the multiple minima problems and more
visibility into the optimization process to gain confidence in
the results.
MDO has a potential as a crucial component in the Concurrent Engineering.
The best way to introduce MDO is by incremental changes.
Trajectory optimization is a good example of an application where
optimization is used because no other means would do.
| NAME | ORGANIZATION |
| Dr. Jaroslaw Sobieski, Chairman | NASA Langley Research Center, MS 246
Hampton, VA 23681-0001 |
| Mr. Jan Aase | Engineering Computing Systems Technology
MD 24043 General Electric 1000 Western Ave. Lynn, MA 01910 |
| Dr. Frank Abdi | Rockwell International
P.O. Box 92098 201 N. Douglas St. #GB15 El Segundo, CA 90009 |
| Dr. Ramesh K. Agarwal | McDonnell Douglas Research Laboratories
Dept. 222/B.110 P.O. Box 516; MC 1111041 St. Louis, MO 63017 |
| Dr. Todd J. Beltracchi | The Aerospace Corporation
P.O. Box 92957 Los Angeles, CA 90009-2957 |
| Dr. Laszlo Berke | NASA Lewis Research Center
21000 Brookpark Rd. Cleveland, OH 44135 |
| Mr. Christopher Borland | Boeing Commercial Airplane Group
P.O. Box 3707; MS 7H-94 Seattle, WA 98124 |
| Dr. Kyung K. Choi | College of Engineering
The University of Iowa Iowa City, IA 52242 |
| Mr. Robert D. Consoli | General Dynamics Fort Worth Div.
Dept. 0635 P.O. Box 748 MZ 2872 Ft. Worth, TX 76101 |
| Dr. Evin Cramer | Boeing Comp. Services
P.O. Box 24346, M/S 7L-21 Seattle, WA 98124-0346 |
| Mr. Alan J. Dodd | Douglas Aircraft Co.
McDonnell Douglas Co. 3855 Lakewood B., M/S 18-86 Long Beach, CA 90846 |
| Prof. George S. Dulikravich | Aerospace Engineering Dept.
233 Hammond Bldg. The Pennsylvania State University University Park, PA 16802 |
| Mr. George C. Greene | Fluid Mechanics Div., MS 163
NASA Langley Research Center Hampton, VA 23665 |
| Dr. Zafer Gurdal | Engineering Sc. and Mechanics Dept.
Virginia Polytechnic Institute Blacksburg, VA 24061 |
| Dr. Prabhat Hajela | Dept. of Mechanical Engineering
Aeronautical Eng. and Mechanics 5020 Jonsson Eng. Ctr. Rensselaer Polytechnic Institute Troy, NY 12180 |
| Dr. Wayne Hallman | The Aerospace Corporation
P.O. Box 92457 Los Angeles, CA 90009 |
| Dr. K. Scott Hunziker | Boeing Aerospace
P. O. Box 3999, M/S 82-97 Seattle, WA 98124-2499 |
| Dr. Erwin H. Johnson | MacNeal Schwendler Co.
815 Colorado Blvd. Los Angeles, CA 90041 |
| Dr. Ilan Kroo | Dept. of Aeronautics and Astronautics
Stanford University Stanford, CA 94305 |
| Mr. Michael Love | General Dynamics Fort Worth Div.
P. O. Box 748, MZ 2824 Ft. Worth, TX 76101 |
| Dr. John K. Lytle | NASA Lewis Research Center, MS AAC-1
21000 Brookpark Rd. Cleveland, OH 44135 |
| Mr. Philip Mason | Grumman Aircraft Systems Div.
MS B43/35 Bethpage, NY 11714 |
| Dr. Hirokazu Miura | System Analysis Br.
NASA Ames Research Center, MS 237-11 Moffett Field, CA 94035 |
| Mr. Douglas Neill | Northrop Aircraft Div.
Dept. 3854/82, 1 Northrop Ave. Hawthorne, CA 90250 |
| Mr. Larry G. Niedling | McDonnell Aircraft Co.
P. O. Box 516, M/C 03412 80 St. Louis, MO 63166 |
| Ms. Beth Paul | General Dynamics Ft. Worth Div.
P. O. Box 748, MZ 2208 Ft. Worth, TX 76101 |
| Dr. Nick Radovcich | Lockheed Aeronautical Systems Co.
Dept. 76-12, Bldg. 63GE, Plant A-1 P. O. Box 551 Burbank, CA 91520 |
| Mr. Bruce A. Rommel | Douglas Aircraft Co.
McDonnell-Douglas Corp. M/S 18-86 Long Beach, CA 90846 |
| Dr. Vijaya Shankar | Rockwell Int'l. Science Center\
P. O. Box 1085 Camino del Rios Thousand Oaks, CA 91360 |
| Dr. Daniel P. Schrage | School of Aerospace Engineering
Georgia Institute of Technology Atlanta, TA 30332 |
| Mr. Otto Sensburg | MBB Ottobrunn
P. O. Box 80 11 60 8000 Munich 80 Germany |
| Mr. J. Tulinius | Rockwell International Corp.
North American Aerospace Oper. 011 GC02 P. O. Bhox 92098 201 N. Douglas St. El Segundo, CA 90009 |
| Dr. Gary Vanderplaats | VMA Engineering
5960 Mandarin Ave., Suite F Goleta, CA 93117 |
| Dr. Vipperla Venkayya | Air Force Wright Research & Dev. Center
FIBR Wright-Patterson AFB, OH 45433-6553 |
| Dr. B. P. Wang | University of Texas at Arlington
P. O. Box 19023 Arlington, TX 76019 |
| Mr. John W. Hayn | McDonnell-Douglas Missile Co.
P. O. Box 516, M/C 270 0120 St. Louis, MO 63166 |
