Title: SystemsofSystems Analysis of Ballistic Missile Defense Architecture Effectiveness through Surrogate
1Systems-of-Systems Analysis of Ballistic Missile
Defense Architecture Effectiveness through
Surrogate Modeling and Simulation
SysCon 2008 IEEE International Systems
Conference Montreal, Canada April 710, 2008
Paul Miceli, W. Dale Blair, Phil West Georgia
Tech Research Institute Georgia Institute of
Technology Atlanta, Georgia 30332-0801
- Tommer Ender, Ryan Leurck,
- Brian Weaver, Dimitri Mavris
- Aerospace Systems Design Laboratory
- School of Aerospace Engineering
- Georgia Institute of Technology
- Atlanta, GA 30332-0150
2Introduction
- Realizing an effective BMD system has been a goal
of U.S. military planners since the first German
V-2 rockets began falling on London during WWII - Emergence of new nuclear threats
- North Korean ballistic missile tests in summer of
2006 - Suspected ongoing nuclear weapons development
program in Iran
An integrated, layered Ballistic Missile Defense
System to defend the United States, its deployed
forces, allies and friends from ballistic
missiles of all ranges and in all phases of
flight. -MDA
3Challenges
- No historical database of working BMD systems
from which designers can generate trends or
rules of thumb to help reduce design cycle time - Scope of the problem limits physical testing
- Full-scale test of realistic threat scenario
expensive - Modeling and simulation (MS) required
- Modeling of entire BMD scenario too
computationally expensive to be useful in
earliest stages of design
Emerging techniques for quickly evaluating the
performance of complex systems-of-systems can be
used to study BMD
4Architecture Diagram
Analysis codes for each component are extremely
complex
Even if all tools could be successfully
integrated, runtime would be prohibitive
www.mda.mil
5Systems-of-Systems Architecture Analysis
- Systems-of-Systems
- Elements that are operationally and managerially
independent 2 - Collaborating only through information exchange
- Capability of the integrated whole to produce
results greater than sum of the individual
components -
-
BMD is an SoS problem targets, interceptors,
radars, and launch platforms evolve independent
of one another
- Architecture Analysis
- SoS architecture acts as a framework that directs
the interaction of components with their
environment, data management, communication, and
resource allocation 3 - The system-of-systems architecture defines the
interfaces and composition which guides its
implementation and evolution ?4
Allocation of functionality to components and
inter-component interaction, rather than the
internal workings of individual components
6Current Approaches for BMD Architecture Analysis
- Wilkening BMD effectiveness determined by
modeling the integrated system as a series of
Bernoulli trials 5 - Required numbers of interceptors needed as a
function of key parameters, target properties and
rules of engagement - Ben-Asher applying SoS design principles when
defining architecture enhances the probability of
creating successful BMD systems 6 - Limited to defining responsibility sharing rules
and interfaces between system level components - Parnell methods should enable interaction
between decision makers, operators, and
developers, through real-time what-if scenarios
7
Common Issue Rapid architecture analysis of the
entire BMD problem sacrifices MS accuracy
7Functional Decomposition
Search/Scan
- A series of battle management elements that
execute in a serial process - Comparable to standard kill chain
- Execution of any element is conditional on the
success of the previous element
Detect
Report
Locate
Track
Develop track
Report
Associate
Sensor Fusion
Fuse
BMD Battle Management
Report
Prioritize Threats
Fire Control
Identify Candidate Platforms
Develop Fire Control Solution
Launch
Update target/track
Engage
Fly to intercept
Activate seeker
Discriminate target
Endgame
Kill Assessment
NOTE No claim is made that the complete BMD
problem is represented in this study
Determine Kill
Report
8Proof-of-Concept Scenario
- Used to scope the elements of the problem
decomposition that are necessary to be studied in
detail
- 4 Threats
- Launched near simultaneously
- 2 Ships
- Stationed within an operational area
- Sensor/shooter platforms
- 1 Land Based Interceptor
- Fixed location
- Sensor/shooter platform
Four near-simultaneous Threat Launches
Operational Area for two battleships
Land Based Interceptor
9Surrogate Models
- Surrogate models enable rapid manipulation of
any modeling and simulation tools - Equation based regressions of complex codes
- Negligible loss in accuracy of original tools
- Can be executed in fractions of a second instead
of hours or days - On-the-fly tradeoffs yield results that otherwise
may not have been discovered - Enables decision making across a
systems-of-systems hierarchy
Bringing Modeling Simulation Forward in the
Decision Making Process
10Neural Network Surrogate Models
- Any regression model must make assumptions as to
the behavior of a measured response and accept a
certain amount of error - i.e. Response Y roughly varies linearly with
variables X1 and X2 - Polynomial based Response Surface methods are
proven - How does one create regressions in which a
functional form behavior assumption is not
possible?
Design Space Behavior
Regression Functional Form
How do we fit this?
Equation Form
Solve for
Y mxb
m, b
Y ax2bxc
a, b, c
Y ax3 bx2 cx d
a, b, c, d
Neural Network
Linear Fit
Quadratic Fit
Cubic Fit
Neural Networks are emerging as a useful way of
creating highly nonlinear regression models
11Neural Network Surrogate Models
Neural Network Structure
- A Neural Network is a computational
architecture based on the design of the
interconnections of neurons in our brains - The elements of Neural Networks were inspired by
biological nervous systems, in which the
connections between elements determines the
network function - It is a set of nonlinear equations that predict
output variables from a set of given input
variables using layers of linear regressions and
S-shaped logistic functions
Input Layer
Hidden Layer
Output Layer
H1
X1
H2
X2
H3
Y1
H4
X3
Single Response
H5
Xn
Pattern of Connections Found by Training
Hn
of Hidden Nodes Defined by Optimization
12Modeling and Simulation (MS)
- Goal of BMD simulation is to model protection of
some area from incoming targets by engaging them
with interceptors - Consider MS of all elements within the kill
chain
Interceptors
Scenarios
Sensors
Tracking
Sensor Fusion
Battle Management
Sensor 1
T 1
Sensor 2
T 2
EVALUATE Intercept. Success
.
.
.
.
.
.
Sensor n
T n
Performance Metrics
- Defended Area
- Leakers / Interceptors . .
13Computational Complexity of BMD Modeling and
Simulation
Due to computational complexity BMD simulation is
very computationally intensive and time consuming
- Run Time Example
- Consider 9 TBM Threats, 16 Sensors all in fixed
locations 1 set of Monte Carlo (MC) runs 10
Hours - Parameter Variation
- Hold launch points constant, vary impact among 5
different locations - Move each radar though 6 locations
- Corresponding Run Time
- 10hrs 59 166 3.3x1015 hrs gt 37,000,000
years - (if run on 1,000 parallel machines)
- Many other system parameters of interest remain
14Target Detection and Tracking
- Due to computational complexity most tools for
studying BMD system performance are either - Truth Based
- Any target in a systems field of view (FOV) if
presumed to be under track - Ignores reality, since the BMD system performance
is highly dependent on ability of system to
detect and track targets - Performance Prediction Based
- Based on sensor and tracking parameters, however
- Do not require Monte Carlo Simulations
- Have fast run times
- Typically make many assumptions
Performance Prediction Target Detection
Tracking Techniques are Used for this Study
15Sensor Model
- Two Phased Array Radars
- Notional ground based X-band radar
- Notional ship based S-band radar with four array
faces - Assumptions
- Perfect resolution so that merged measurements
not an issue - No false alarms
- Probability of detection is 1 given a sufficient
signal-to-noise ratio
- Sensor Parameters
- Sensor / Target Geometry
- Monopulse Parameters
- Sensor / Target Geometry
- Off bore-sight steer angle
Local Tracker Performance Prediction
Compute Measurement Covariance
Compute SNR
16Local Tracker Performance Prediction
- Based on a measurement covariance and track
filter parameters, tracker performance is
expressed as a noise only covariance and maneuver
lag - Assumptions
- Perfect measurement to track association
- Steady State
- Constant update rate
- Critical in the future to relax assumption of
perfect measurement to track association - Closely spaced objects a BIG deal in BMD
17System Track Performance Prediction
- Centralized Multi-Sensor Track Fusion
- Fuse all local tracks at a centralized location
- Performance prediction techniques used to assess
the quality of the fused tracks - Reporting Responsibility
- System track selected as the best of all local
tracks for a given object under tracks - Used in Link-16
- Architecture level effects?
Two network level tracking schemes considered
18Engagement Planner Analysis
- Define the battle space as the set of all
possible launch solutions - Computed for each interceptor and target pair for
all times at which, target is presumed to be in
coast phase
Predicted Intercept Point 1
Predict target trajectory using ballistic
extrapolation
X
Predicted Intercept Point N
X
Fly-out fan trajectory
19Engagement Planner Analysis
- For each possible intercept as determined by
fly-out fans - Compare the maneuverability of the interceptor to
a critical threshold - Threshold a function of predicted system track
accuracy - Compare at three key decision points
- Interceptor Launch Time
- Interceptor Burnout Time
- Interceptor Seeker Activation
- A launch solution is possible if key decision
points pass - Compute Probability of Kill (PK)
20Probability of Kill
Pib
PK is a function of track covariance and
interceptor seeker field of view (FOV)
Align track covariance with interceptor body frame
Pecef
U
V
W
21Probability of Kill
Pib
Compute distance in each cross range coordinate
DcrV
V
Area under curve is Gaussian CDF
f Seeker Field of View
W
22Fire Control
- Single engagement manager supervising
interceptor-target assignments over all platforms - Utilized a shoot early strategy
- Threats were engaged when a shot was available
with a Pk exceeded a threshold - Time dependent threshold used to instill
confidence in each chosen shot while being more
accepting of low Pk as the threat nears - Shots constrained by
- Availability of platforms
- Interceptors currently engaging threat
- Pk for a given shot
23Results Placement Trade
- Surrogate models of MS used for Monte Carlo
system performance assessment of ship operational
area - 10,000 Ship/Sensor2 (S2) locations are considered
for a given single Ship/Sensor 1 (S1) location - Ground Based Interceptor/Sensor (S3) fixed all
times - System performance, probability of zero leakers
(P0L), provides operational contours
Expanded Operational Area
New S1 Location
24Results Tracking Method Trade
- With the current Reporting Responsibility
architecture, it is not possible to get a high
confidence of success - Holding the interceptor and fire control systems
constant, surrogates are used to explore
different sensor architectures
Reporting Responsibility
Track Fusion
P0L Results Min 0.00 Max 0.60 Avg 0.11
P0L Results Min 0.53 Max 0.87 Avg 0.79
Track fusion vastly improves P0L in terms of
capability and robustness
25Results Number of Sensors Trade
- Track fusion results show the system performance
is not sensitive to S2 location - Surrogates are used to explore the number of
sensors used
- Models were generated with the forward deployed
land based sensor removed
- Two ship track fusion architecture yields
- Sizeable operational area
- Improved confidence over three platform reporting
responsibility - Note Two ship track fusion requires more
communications bandwidth and different processing
Probability of Zero Leakers (P0L)
2 Platform Track Fusion 0.00 0.83 0.20
3 Platform Report. Responsibility 0.00 0.60 0.11
Min Max Avg
26Conclusions
- Surrogate models used to greatly improve runtime
of large system-of-systems simulation with
negligible loss in fidelity - Enabled real-time system-of-systems level
architectural trade studies utilizing accuracy of
high-fidelity MS tools - Asset placement trade
- Tracking method trade
- Number of sensors trade
- Plan to increase fidelity of various modeling and
simulation of current BMD elements
27 28Engagement Planner Analysis
Condition 1 for launch
Position Covariance
Track covariance extrapolated to predicted
intercept
Track covariance at interceptor launch
29Engagement Planner Analysis
Condition 2 for launch
Position Covariance
Track covariance at interceptor burnout
Track covariance extrapolated to predicted
intercept
30Engagement Planner Analysis
Condition 3 for launch
Position Covariance
Track covariance extrapolated to predicted
intercept
31References
- Myers, R. H. and Montgomery, D. C., Response
Surface Methodology Process and Product
Optimization Using Designed Experiments. New
York John Wiley Sons, Inc, 1995. - Maier, M. W., Architecting Principles for
Systems of Systems, in Proceedings of the Sixth
Annual International Symposium, International
Council on Systems Engineering, Boston, MA, 1996. - Clements, P., Krut, R., Morris, E., and Wallnau,
K., The Gadfly An Approach to
Architectural-Level System Comprehension, Fourth
IEEE Workshop on Program Comprehension, Berlin,
March, 1996. - Chen, P., Gori, R., and Pozgay, A., Systems and
Capability Relation Management in Defense
Systems-of-System Context, , Proceedings of The
Ninth International Command and Control Research
and Technology Symposium, Copenhagen, Denmark,
2004. - Wilkening, D.A., A Simple Model for Calculating
Ballistic Missile Defense Effectiveness, Science
Global Security, Volume 82, 1999, pp183-215. - Ben-Asher, J.Z., Systems Engineering Aspects in
Theatre Missile Defense-Design Principles and a
Case Study, Systems Engineering, Vol. 7, No. 2,
2004. pp186-194. - Parnell, G.S., Metzger, R.E., Merrick, J.,
Eilers, R., Multiobjective Decision Analysis of
Theater Missile Defense Architectures, Systems
Engineering, Vol. 4, No. 1, 2001, pp24-34. - Ball, R. E., The Fundamentals of Aircraft Combat
Survivability Analysis and Design. AIAA, 2003. - Ender, T.R., A Top-Down, Hierarchical,
System-of-Systems Approach to the Design of an
Air Defense Weapon. PhD thesis, Georgia
Institute of Technology, 2006. - Bar-Shalom, Y., Li, X.R, and Kirubarajan, T.
Estimation with Applications to Tracking and
Navigation, John Wiley and Sons, Inc. New York,
NY, 2001. - Bar-Shalom, Y., and Blair, W.D., editors,
Multitarget-Multisensor Tracking Applications
and Advances III, Norwood Massachusetts Artech
House (2000). - Blair, W.D. Fixed-Gain, Two-Stage Estimators for
Tracking Manuevering Targets, IEEE Transactions
On Aerospace and Electronic Systems, July 1993,
pp. 1004 1014 - Blair, W.D. and Miceli, P.A., Performance
Prediction of Multisensor Tracking Systems for
Maneuvering Targets, submitted to the Journal
for Advances in Information Fusion, April 2008.
32Presenter Instructions (IEEE)
- Oral presentation consists of 45 minutes for
presentation of the problem, the background, the
innovative approach, the new results, the
comparative evaluation and questions from
attendees. Acceptable formats for presentations
are MS PowerPoint 2003. - To perform the presentation, the following
equipment will be available in the conference
room - Laptop computer (with MS Windows XP-PRO operating
system and PowerPoint 2003) and LCD projector. - Please bring your presentation on a memory stick
to upload on the laptop provided prior to your
session. - The presenter should prepare a reasonable number
of electronic slides, so as not to exceed the 45
minute time. Typically, 1 slide is presented in 1
minute. Additional slides could be prepared to
support possible answers to attendees' questions.
- Each slide should not be overcrowded by text and
graphics. Too much text should be avoided slides
should support the presentation, they should not
be simply read by the presenter. - Graphics should be designed to help in the
overall communications, to illustrate key points
and illucidate the basic ideas. Avoid excess
graphics as this makes the overall file size
large and can delay slide loading. - Use large fonts so as they are readable without
efforts typically 20 point font is the smallest
that should be used. Headers within the body are
typically 24-pt and slide headers are typically
28 or 32-pt font. Use Arial, Verdana or other
san-serif font for readability.