Title: Predictability Issues in Aircraft Analysis, Design, and Certification Chris L' Pettit, Ph'D', P'E' M
1Predictability Issues in Aircraft Analysis,
Design, and Certification Chris L. Pettit,
Ph.D., P.E. Multidisciplinary Technologies
CenterAir Vehicles Directorate, Air Force
Research Laboratory
JHU Predictability Workshop, November 13-14, 2003
2About This Presentation
- Organizers goal Synthesize a template for
quantitative processes related to predictability
and UQ - My goals as moderator Define context and
highlight key questions to motivate group
discussion - Describe prediction problems being confronted in
(military) aircraft design and certification,
including - Nonlinear, multidisciplinary, and multi-scale
problems - Prediction difficulties that limit the
performance and health of current systems and the
development of future systems - Promote discussion and feedback on key topics
- Definitions of predictability and
predictability-aware models - How to assess predictability and what to do with
it - Error vs. uncertainty
- Roles of testing during various phases of design
and life cycle - Role of predictability assessment in aircraft
systems engineering and decision-making (What is
the risk associated with low predictability?) - Current impediments to predictability
- Benefits of higher predictability (e.g., cost,
performance, safety)
3About This Presentation (cont)
- Ill try to avoid injecting unnecessary bias into
what often are controversial philosophical issues - I hope to learn more from you than you will from
me - But, I will assume
- We want to measure our ability to model complex
systems - We are uncertain about all processes
- Models are not reality
- Natural and man-made physical systems are never
deterministic - Assessing predictability requires uncertainty
quantification (UQ) - Each model has a limited range of validity
- Model validation ultimately depends on UQ and
model usage - Predictability ? Model Validity
- Is the reverse true?
- Physics-based models promote predictability
assessment - Error estimators, safer extrapolation, etc.
- Context does matter military aircraft prediction
is part of the DoD acquisition process ? Models
should help to assess system-level risks!
4Some tough prediction problems designers and
analysts are facing
5Prediction-Critical Disciplines for Current and
Future Aircraft Systems
These are disciplines that lead to severe
performance restrictions, high required margins,
and re-designs
- Extreme environments (e.g., thermoacoustic loads)
- Nonlinear aeroelasticity
- Flow control and mixing
- Signature reduction
- Radar cross-section (RCS)
- Thermal
- Structural integrity
- Fatigue, fracture, corrosion, delamination,
battle damage, etc. - Strongly dependent on other disciplines and usage
for loads - Sensitivity to manufacturing tolerances
- Structural instability
- Others??? (e.g., dynamics of UAV swarms, human
behavior)
6Common Complicating Factors in Prediction
- Prediction-critical phenomena commonly involve
- complex processes that span multiple spatial
and temporal scales ? At which scales can we be
predictive? - nonlinear processes
- multidisciplinary interactions
- relatively high epistemic and aleatory
uncertainty - low observability in experiments and tests
- sensitivity to BCs and ICs
Each of these factors complicates our
attempts to predict impedes our efforts to
assess predictability
7Current and Expected Practical Prediction
Challenges (1/3)
- Low acceptance of predictive ability
- Especially for safety-critical and multi-scale
phenomena - Model validation is a low priority
- Risk assessment is not trusted
- Accelerated testing requires
- More dependable predictions
- Less subjective risk estimation
- Model and test integration
- Nonlinear systems can be very sensitive to
variability in systems properties, loads, and
BCs - Bifurcations
- Hard to model in complex systems
8Current and Expected Practical Prediction
Challenges (2/3)
- Non-robust optima in aeroelastic tailoring and
laminar flow wings - Manufacturing variability
- Off-design conditions
- Non-traditional design concepts, and highly
variable or extreme operating environments - Little historical basis for assessing loads and
sensitivities - Difficult to estimate risks of new technology or
design concepts (e.g., TRL assessment) - Untapped potential because of the low
predictability??? - Are dated safety reqs holding back existing and
new technologies?
9Current and Expected Practical Prediction
Challenges (3/3)
- Designer materials and non-traditional structures
- Prediction of properties across length scales
- Ensuring adequate performance in non-ideal
conditions - Avoiding unintended failure modes
- Multi-functional structures and systems
integration - Structurally integrated (i.e., load-bearing)
antennas - Distributed control surfaces and shape control
- Optimization of control laws in multiple flight
regimes - Load redistribution for non-aerodynamic or
non-structural purposes (e.g., antenna pointing
or RCS management)? - Integrated vehicle health management (IVHM)
systems - Data fusion and model-based sensor placement
optimization - On-line modification of control laws for loads
management - Design of self-healing materials
- Airframe-propulsion integration in hypersonic
vehicles - System-level performance metrics
- Defining trade-offs given multiple energy flow
paths - Multiple performance modes requires
multidisciplinary models
10Predictability in the context of aircraft design
and certification
11How the Tough Problems Affect Processes and
Frameworks
- Current aircraft systems already stress design
and certification methods to (or beyond?) their
practical limits - Unique design concepts suggest increased
importance of nonlinear multidisciplinary physics
clearly beyond capability of current design tools
and certification processes - Physical and computational complexity of
nonlinear multidisciplinary models obscures the
propagation of uncertainty through networks of
models - Difficult to dependably assess sensitivities and
risks w/o a clear UQ process that is consistently
implemented - For airframes This has resulted in a
process-centric approach to risk management
instead of a knowledge-centric approach - This is untenable for future Air Force needs
12Multidisciplinary Problems
- Very hard to predict and validate
- Multi-scale, nonlinear physics
- The correct uncertainty model often depends on
physics modeling choices and measurement
limitations - e.g., stress FE models vs. dynamics FE models
- Highly variable operating environments, loads,
and material properties - Complicated and expensive tests
- Crucial to the success and safety of
high-performance military aircraft - ? Computational multidisciplinary analyses are
always suspect, as is any resulting risk
prediction
13Predictability in Systems Engineering (SE)
- Prediction must be performed and assessed in the
context of systems engineering - Purpose of SE manage system-level risks from
cradle to grave - Risk results from uncertainty and error
- Risk management demands good data and good
predictions - ? Risk management requires predictability
assessment - SE entails a risk allocation or flow-down from
program level to system, sub-system, and
component levels - Usually implicit and qualitative for complex
systems - This flow-down parallels a similarly implicit
flow-down of uncertainty in multidisciplinary
design problems - Modeling and data-gathering decisions
automatically allocate uncertainty and error to
constituent analyses - Uncertainty and error budgets are never described
explicitly and are extremely difficult to
quantify - ? Predictability assessment ultimately needs
UQ
14Uncertainty Flow-Down
- How much uncertainty can be tolerated in the
top-level prediction of a multidisciplinary
process? - How much uncertainty can be attributed to each
sub-discipline in the network of models that
comprise the multidisciplinary analysis? - Must address epistemic and aleatory sources
- How much uncertainty can be tolerated in each
sub-discipline analysis? - How do the modeled physics amplify input
uncertainty? - What test, computer, and training resources must
be invested to assess and control the uncertainty
in each sub-discipline?
Do these work for error flow-down also? Do these
really help in assessing predictability?
15Prediction and Information-Management Tools
- Design and test cycles of military aircraft now
exceed 20 years! - Many airframe designers now work on only a few
new aircraft programs during their entire career - Opportunities to gain practical experience are
extremely limited - Can no longer depend on old-timers as the
primary storehouses of corporate knowledge - Retirements and overwhelming demands on their
time - Even they may not have insight into
non-traditional problems - Worse yet They can be nay-sayers
- Analyses used to support design decisions may be
obsolete by the time the aircraft is certified - DoD Acquisition Reform Mandated evolutionary
acquisition and spiral development processes
institute definite needs for more complete
knowledge to support future upgrades - How can prediction frameworks be structured to
overcome these difficulties???
16Closing Remarks about Aircraft Predictability
- Predictability must be assessed in terms of which
questions are being answered by the model - Prediction-critical aircraft phenomena share many
complicating characteristics - Ability to be predictive and to assess
predictability is fundamental to future military
aircraft systems and acquisition processes - UQ and error estimation are fundamental to
predictability - Predictability depends as much on the practical
details of modeling and testing process (e.g.,
best practices, ability to measure key data) as
it does on theory
17Whats next?
18Breakout Group Plan of Action
- I will present several suggested topics of
discussion - Summarize first, then cover each separately in
detail - Each topic addresses some of the concerns Ive
discussed already - Try to step through the topics one-by-one for
group discussion - We have little time
- Please try to confine your remarks to the
question at hand - I encourage open discussion, but I will press
ahead if we do not move through the topics
quickly enough. Please dont be insulted if I
abruptly terminate a portion of the discussion. - Remember Our ultimate goal is to begin
developing a template for aircraft predictability
assessment in the context of uncertainty and error
19Suggested Topics of Discussion
20Suggested Topics of Discussion
- What is the working definition of predictability
in the context of aircraft analysis, design, and
certification? - What is the current state of UQ and
predictability awareness for aircraft? - How can aircraft predictability be assessed
objectively? - What are the dominant modeling, testing, and
validation challenges that impede aircraft
predictability? - What content must a predictability-aware model
of a complex aircraft system offer? - What new things could be accomplished in
aircraft analysis if predictability were
substantially improved?
21Topic 1
- What is the working definition of predictability
in the context of aircraft analysis, design, and
certification? - Does it differ substantially between the Critical
Disciplines cited earlier? - Do we need to clarify the relationship between
predictability, model validity, error estimation,
and UQ? - I cant define what it means to be predictive,
but I know it when I see it.
22Topic 2
- What is the current state of UQ and
predictability awareness for aircraft? - Does it differ substantially between the Critical
Disciplines? - Research vs. practice?
- Do decision-makers place sufficient priority on
predictability assessment?
23Topic 3 (1/2)
- How can predictability be assessed objectively?
- What are the appropriate metrics? Is model
validity truly a prerequisite? - What is the role of experimental evidence in
understanding, measuring, and controlling
predictability? - How is uncertainty related to error estimation?
- Numerical error vs. statistical error?
- Is a converged deterministic grid automatically
good for UQ? - How should the error and uncertainty budgets be
decomposed to clarify predictability assessment? - Global vs. local measures of predictability?
- Throughout the design parameter space?
- Throughout the spatio-temporal extent of a given
design and its model? - Scaling issues in comparing tests and models?
- Will the common scale factors (Re, Fr, etc.)
remain the most important as non-traditional
designs are developed? Note This is already an
issue for aeroelastic wind-tunnel models.
24Topic 3 (2/2)
- Are acceptable confidence measures available for
error estimates? - What is their nature (e.g., fuzzy vs. subjective
probability?) - Is there agreement on how to combine component-
and discipline-level error estimates to obtain
system-level error estimates? - How should these be communicated to
decision-makers?
25Topic 4 (1/2)
- What are the dominant modeling, testing, and
validation challenges that impede aircraft
predictability? - Where in the prediction chain do the limitations
enter? - Availability of accurate input data and its
variability? (e.g., constitutive properties,
geometry, etc.) - A priori knowledge of input errors/uncertainty
and their consequences? - Math models? Could include unresolved physics
- Algorithmic implementation of math models? This
could include discipline coupling. - Numerical sensitivity (grid and time step,
convergence criteria, etc.) - Short-term vs. long-term accuracy? Dependable
error estimators? - Post-processing and interpretation? Model
validation and integration with testing?
Availability of dependable test data? - Can we trade some full-scale tests for more
coupon and component tests to improve UQ and
error estimation? - How are the challenges shaped by the push to
reduce test resources and streamline
certification decision-making?
26Topic 4 (2/2)
- Which important measurements cannot be made with
current capabilities? - Are these limitations controlled by physics,
technology, cost, resource prioritization, or
something else? - How can validation plans be adjusted to mitigate
these limitations? - Given that aircraft normally admit some testing
throughout the design process, how should these
test resources be allocated to estimate model
errors and uncertainty? - Other impediments to predictability assessment
not mentioned yet???
27Topic 5
- What content must a predictability-aware model
of a complex system offer? - How does this depend on the purpose of the model?
- Who will use it? When? Why?
- How can information and high-fidelity analysis
frameworks be structured to promote
predictability? - Which types of prediction difficulties are best
addressed through process structuring and
control? - How can frameworks be used to promote
communication between analysts and test personnel
in estimating predictability? - Should the model carry supporting data in
parallel to support predictability assessment? - What about enforced recording of modeling
assumptions and decisions? - Multiple spatial and temporal scales?
28Topic 6
- What new things could be accomplished in
aircraft analysis if predictability were
substantially improved? - How is aircraft performance predictability-limited
? - Reduce required margins and safety factors?
- How much of a safety factor is allocated to cover
modeling errors and missing info vs. inherent
variability? - Is a system-level risk or uncertainty budget a
practical concept? - Can it be allocated rationally to components or
modeling disciplines? - Should predictability goals be tied to different
stages in the design and certification process? - Can predictability become a trade-off variable in
the systems engineering process? Is this a
function of the size of the production run?
29Anything else?
30Backup Slides
31Impediments to Reliable Risk Analysis of MD
Aircraft Problems
- System-level risks generally involve
incommensurate types of ignorance whose relative
importance is problem-dependent and
discipline-dependent - No universally accepted way to measure and
combine these types of ignorance consistently - Industry mindset often prefers wrong answers that
come quickly to better answers that take longer - Design process is perceived as a time and
resource sink that must be tolerated in order to
generate revenue downstream - Certification processes automatically biased
toward the technological status quo - Potentially delays transition of beneficial new
structures and materials technologies
32 The Role of Processes and Frameworks
- We need a comprehensive approach to storing
models, traditional analysis results, UQ results,
and any other info used to support design and
certification decisions (e.g., expert opinions) - Must facilitate guided access for future
generations to support - Future expansions of operational capabilities
- Life-extension programs
- Insight into the sources and solutions of
unexpected problems - Should also promote informative modeling and
analysis practice (including UQ) by requesting - Key inputs and outputs, including their uncertain
aspects - Documentation of modeling decisions
33Our Perspective
- UQ-based analyses are needed to help reveal
unexpected failure modes and to assess their risk - We already do a reasonable job of preventing most
well-known structural failure modes in
traditional designs - Could be critical for non-traditional designs
- UQ-centric processes promote maximum payoff from
models and tests at all scales (e.g.,
coupon-level to full-scale) - Motivation behind test-planning should be
transformed to emphasize model validation in
addition to (or instead of?) certification
criteria - This will require substantial modification of
traditional RD and program funding profiles - Allocate additional funds during conceptual and
preliminary design stages to support additional
data gathering and analysis activities ? Need to
fill the pool of knowledge as early as possible!
34What Should Our Goals Be?
- USAF needs to increase reliance on
multidisciplinary analysis earlier in the design
process - Detect genuinely avoidable problems before
full-scale ground and flight tests - Achieve operational capabilities and efficiencies
by enabling access to portions of the design
space that are precluded by current certification
requirements and precautionary biases - Ideal outcome Dependable quantification of
technical and performance risk early on lead to - Informed assessment of competing technologies
- Accelerated insertion of new material,
manufacturing, and assembly processes - Proactive prevention of problems instead of
compromise fixes after problems are uncovered
during testing
35Systems Engineering Concepts
- System An integrated composite of people,
products, and processes that provide a capability
or satisfy a stated need or objective - Systems Engineering (SE) An interdisciplinary
engineering management process that evolves and
verifies an integrated, life-cycle balanced set
of system solutions that satisfy customer needs - Our premise The goal of SE is to make informed
decisions that efficiently mitigate risks while
meeting goals - Every goal induces risk!
Risks results from uncertainty! ? UQ should be
part of SE
Systems Engineering Handbook, DAU Press, 2000.
36Uncertainty and Systems Engineering for
Aeroelasticity
37Airframe Certification (1/2)
- Certification The end result of a structured
process for identifying and managing risk from
conception to regular operation - Current processes
- Little reliance on analysis for risk assessment
- Fail to promote interaction between tests and
analyses - Inadequate for future materials, technologies,
and design concepts - Result? Structures certified through
safety-factor design and expensive building
block tests - Additional spent to certify repairs (e.g.,
fatigue hot spots) and operational modifications
(e.g., aeroelastic stability with new external
stores) - Even certified airframes still have many
unexpected problems - How can we learn tomorrow what were not learning
today???
38Airframe Certification (2/2)
- ASIP USAF certification process for structural
integrity - Reasonably successful but several shortcomings
- Time-consuming and manpower intensive
- Dependent on historical database
- Risk assessment is too qualitative and subjective
- USAF striving to increase reliance on analysis in
airframe certification through - Higher-fidelity modeling earlier in design
process - Uncertainty quantification (UQ) for risk analysis
- Verification and validation of models
- Streamlining and expanding knowledge generation
and management processes - Why?
- Increase safety and likelihood of achieving
performance goals - Save time and money by reducing or eliminating
some tests and accelerating iterative design
processes - Avoid costly changes late in design cycle
- Will these benefits actually be realized? TBD
39Prediction and Information-Management Tools (2/2)
- Need tools that actively promote the gathering
and retrieval of relevant information - Knowledge-bases for capturing and accessing
- Conceptual design support info (e.g., historical
requirements and capabilities) - Concept maps and influence diagrams for
system-level interactions - Product-centric, object-oriented design
environments that capture the methods operating
on each product - Could include enforced documentation of modeling
decisions - Also need tools that support consistent and
rational data fusion, inference, risk assessment,
and decision-making - Automated best practices and guided
model-checking - Measures of confidence associated with expert
opinions - Consistent model validation processes
402-DOF Airfoil LCO Problem Description
- Subcritical Hopf bifurcation
- 5th-order pitch spring
- k3 lt 0 ? destabilizing
- MCS on a(t 0), k3, k5
- 4,000 realizations at each reduced velocity
- Incompressible, unsteady aero (Jones
approximation)
MCS
41Current Issues in Uncertainty Quantification for
Airframes
42Context for Identifying Research Challenges
Analysis is a tool to support decision making in
design and certification, which is a process of
managing risk while trying to achieve performance
goals
43Overview of Challenges
- Technical Challenges
- Probably familiar to most researchers
- Non-Technical Challenges
- Often a result of institutional issues
- Non-technical because they cant be resolved by
technical advancement alone - Not always exclusive of technology because
established design and certification practice
often reflect assumed technical capabilities - Our Focus Areas in which targeted research can
lead to success given available computing and
testing methods - No Unobtainium allowed!
44Aerodynamic Uncertainties (1/2)
- Typical modeling issues wont go away, but should
they be re-prioritized for stochastic
considerations? - Domain discretization and approximation of BCs
How much precision is justified given aleatory
uncertainties? - Simulation vs. design
- What is appropriate level of fidelity or
complexity? - Where and how to insert uncertainty models?
- Sensitivity to ICs
- Structure? Flow?
- Importance of non-stationary or extreme gust
loads? - Many assumptions commonly made to work around
uncertainty in atmospheric turbulence - von Karman spectrum and gust length scale are
imposed compromises - Nonlinear instabilities sensitive to level of
disturbance - Extreme gust events are not captured
45Aerodynamic Uncertainties (2/2)
- Stochastic CFD for computational aeroelasticity
- Model problem currently under study 2 DOF
airfoil with polynomial chaos for response - Modeled aero only (Jones approximation)
- Uncertain ka, kh, ICs (a0)
- Which problems would require or benefit from
this? - Subcritical bifurcations ? bimodal response pdf
- Bifurcation sensitive to parametric uncertainty
- Second-moment reliability methods not very
reliable here - Need to know the nature of the bifurcation just
to define what constitutes failure. - Integration with reduced-order solvers?
- Institutional issues or roadblocks?
- Training of analysts?
- Integration with existing design tools and
processes?
46Structural and Other Issues
- Structural damping models
- Perhaps a key factor in observed limit-cycles,
but poorly understood - Which issues dont we consider now but need to if
certification required quantitative risk
estimates of aeroelastic stability and
performance? - More off-design conditions?
- Representation of variable fuel and stores loads?
- Uncertainty in composite lay-ups for aeroelastic
tailoring? - Maybe not for low drag, but what about for
embedded sensors (e.g., Sensorcraft) - Certain people dont want to know about the
uncertainties - Opportunities?
- Active aeroelastic wing built-in risk
mitigation???
47Certification Philosophy (1/3)
- Cert needs to be recognized by all as a
structured dialogue that includes - Designers and analysts
- Test and manufacturing personnel
- Cert Officials and users
- This dialogue establishes perceived levels of
acceptable risk for a given aircraft program - Safety and performance
- Cost and schedule
- Political
- Cert officials havent declared how to use UQ to
support cert decisions
48Certification Philosophy (2/3)
- Trade-off studies suggest much potential for UQ
here - Issues that impede use of risk analysis for
airframes - Current analysis and manufacturing capabilities
- Availability of statistically significant input
data - Background and mindset of decision makers
- Legal and societal perception of quantified risk
- Cost and time of design and cert process is high
but known - Safety factors implicitly cover many sources of
uncertainty - Parametric uncertainty, model errors, non-safety
concerns (e.g., serviceability and performance),
and unknown unknowns - How to allocate these in quantitative risk design
criteria?
49Certification Philosophy (3/3)
- Proposed innovative designs offer many unknowns
w.r.t. current certification procedures - Identification of critical failure modes
- Required testing to ensure safety in these modes
- Required safety factors for UAVs no pilot to
protect - Not yet clear if a risk-informed approach would
be adequate for airframes - Probabilistic safety factors (similar to LRFD
in CE) - Airframe failure modes can be harder to identify
a priori than those of civil structures - Hard to integrate new analysis methods and
account for reduced risk associated with
validated models - Hard to test airframes, but much easier than
testing buildings!
50Other Considerations
- Education and Training
- Aerospace engineers get little training in
probability and none in formal risk analysis - Undergrad curriculum does poor job of discussing
practical failure modes and processes - Management often uninitiated also ? hard sell!
- Widespread high-fidelity analysis will require
more sophistication of designers/analysts - Cost
- Potential cost savings of risk-based cert hard to
estimate - Inadequacies of current cert process often not
evident until after long-term operation - Perhaps the true cost of current design and cert
processes should be recalculated to include
downstream consequences