Research Project / Applications Seminar SYST 798 FINAL REPORT Second Dry-Run 24 April 2008 Team: Tom Hare Ali Raza Brian Boynton Eric Ho Matt Maier Key Sponsor: Dr. Kuo-Chu Chang
Dec 28, 2015
Research Project / Applications SeminarSYST 798
FINAL REPORT Second Dry-Run 24 April 2008
Team: Tom Hare Ali Raza Brian Boynton Eric Ho Matt Maier
Key Sponsor: Dr. Kuo-Chu Chang
What is Biological Sensor Fusion?
“Improve Urban Biological Terrorism Response”
Biological
Sensor
Fusion
Responding to National
Biological Threats
Robust Multi-Tiered Detection
System
Data Integration for Immediate
Application and Response
Sponsor Information
• GMU SEOR: Homeland Security and Military Transformation Lab
• Dr. Kuo-Chu Chang, Professor, GMU
– [email protected]– http://ite.gmu.edu/~kchang/
• Dr. Kathryn Blackmond Laskey, Professor, GMU
– [email protected]– http://ite.gmu.edu/~klaskey/
• Held Sponsor Meetings and Project Demos
– 7 Feb, 20 Feb, 6 Mar, 20 Mar, 3 Apr, 16 Apr
Why is a Biological Attack a Threat ?
“Smallpox has killed more people than any other disease in history, including bubonic plague, at least 300M in the 20th Century.” (U.S. Public Health Service)
Centers for Disease Control (CDC)Listing of Potential Bioterrorism Agents
Category A Category B Category C
- High priority- Easily transmitted from person to
person- High mortality rates- Potential for major public health
impact- Can cause public panic/ social
disruption- Requires special action for public
health preparedness
- Second highest priority- Moderately easy to disseminate- Moderate morbidity and low
mortality rates- Specific diagnostics required- Enhanced disease surveillance
- Third highest priority- Emerging pathogens- Could be engineered for mass
dissemination- Available- Easy production and
dissemination- Potentially high morbidity and
mortality- Major health impact
AnthraxBotulismPlagueSmallpoxTularemiaViral Hemorrhagic Fevers
BrucellosisGlandersMelioidosisPsittacosisQ FeverTyphus FeverViral EncephalitisToxinsFood SafetyWater Safety
NipahHantavirus
Source:http://www.hsarpabaa.com/Solicitations/HSARPA_RA-03-01_Appendices.pdf http://www.hsarpabaa.com/Solicitations/HSARPA_RA-03-01_Body.pdf
Research Conducted
• Technical Paperwork– 60+ articles/papers/books to include 20+ sponsor-
provided technical papers
• Subject Matter Expert (SME) Interviews– Use Case: Chicago – Sensor Technology– Modeling Methods
• Research Areas– Detection: Tiered sensor grid– Fusion: Data Aggregation and Geo-Location– Communication: Epidemic, gossip, and geographic
algorithms– Response: Real-time cordon mapping in changing
environment– Technology: State of the art in 2008 and forthcoming by
2020
Earl W. Zuelke Jr., Deputy Director, Homeland Security & Emergency
Management for the City of Chicago
Photo Courtesy Chicago Police Marine Unit
Objective
• Objectives: – Minimize the time it takes to
inform the public of a biological attack
– Design End-to-End System for constant monitoring of urban environment
• Determine system architecture
• Optimize communication parameters and algorithm usage
• Model usage of current sensor technology
• Architectural products will be incorporated through out the presentation
Operational Concept Graphic (OV-1)
Use Case Scenarios
An individual with infectious Smallpox arrives in Chicago by
car, plane or shipVisual Indicators:• Lesions or rash on person, coughing, feverActions:• Transportation and border staff likely to alert hospitals and confine person
Tier I Sensor in airport and Tier II sensors in arrival/departure zones will detect.
A terrorist sprays aerosolized Smallpox into the air at
Chicago’s Soldier Field Stadium
Visual Indicators:• Minimal, act could be mistaken for a fan with an air hornActions:• Nothing for 7-17 days until Smallpox incubation period is complete, and epidemic outbreaks
Tier I Sensor in stadium and Tier II sensors in parking lot will detect.
A terrorist sprays aerosolized Anthrax in a crowded subway
stationVisual Indicators:• Cloud of dry spores evident• Nearby people would get dust on clothes and shoes Actions:• Maybe none, depending on how obvious actions are• 2001+ people sensitive to powders
Tier II Sensor deployed inside or outside subway will detect.
A terrorist sprays aerosolized Tularemia outside a Chicago
monument
Visual Indicators:• Minimal, act might not be noticed by tourists/ passersbyActions:• Nothing for 3-5 days until Tularemia incubation period is complete, and epidemic outbreaks
Tier II sensors, as they move through vicinity, will detect.
High-Level Requirements
• “Improve Urban Biological Terrorism Response”– Lack of detection and fusion today– Slow response times cost lives– False positives cost money
• Biological Sensor Fusion System Requirements:– The Biological Sensor Fusion Group shall create a representative architecture of a
BioSensor System in an urban environment that will include a model that depicts the fusion of data and establishment of an effective cordon.
– The Biological Sensor Fusion Group shall design a solution to provide for fast Data Fusion to facilitate effective command decision-making.
Operational Activity
Decomposition (OV-5, Node Tree)
Millennium Park, Chicago. Photo Courtesy 80s Forum
BioSensor Fusion Context
• System Context– Within Chicago (Pop. 2.9 mil) there is a
potential for 575,000 deaths or more if unchecked
– Realistically, given fast emergency response, roughly 35+ fatalities would occur
– Exorbitant expense– Current response plans would not allow
for detection or response before 3-4 days
– Our Model will investigate employment of both current, and state of the art technology that will not be put into operation for another 10 years
Anthrax Spores, Photo Courtesy of Wired, 10 October
2001 AP Photo
Smallpox Effects, Photo Courtesy of PBS NOVA Online,
Bioterror, November 2001
Assumptions and Constraints
• Accuracy prioritized over fast detection– False alarms that shut down facilities and
displace people can rival the cost of an actual outbreak ($750+ million)
• Technology Limitations– Tier 1 Sensors require 24 hours to scan the air
within their range, Tier 2 (Mobile) Sensors require 4 hours to make a positive identification, and Tier 3 Sensors can detect a Biological Agent within 2 hours of deployment.
• Biological Attack Versatility– The Sensor Technology Deployed can scan and
positively identify a large number of potential Biological Agents, not just smallpox, so this project can generally be considered a template in modeling a response to any type of Biological attack.
Overview and Summary
Information (AV-1)
System Technology
Tier II: Mobile Ad Hoc Sensors– Deployed in emergency response
vehicles (Emergency, Police, Fire, HAZMAT, etc.)
Example:Biowatch 3 Bioagent
Autonomous Networked Detector (BAND)
Example:General
Dynamics Biological
Agent Warning Sensor (BAWS)
Tier I: Stationary Sensors – Permanent, round-the-clock air-
sampling, building installed indoor and outdoor
Tier III: Stationary Ad Hoc Sensors
– Scattered after a threat is confirmed
– Provide tracking of dispersion
Example:Future
Sensors, Pacific
Northwest National
Laboratory Systems Performance Parameters Matrix (SV-7)
Modeling and Simulation
• Colored Petri Nets Model– Analyzes effectiveness of Small World Communication (6
degrees of separation)– Evaluates Delivery Rate vs. Data Buffer Size
• JAVA Algorithm Model– Models the ad-hoc communication of the sensor network– Evaluates 6 different algorithms
• Epidemic-SI, Epidemic-SIS, Epidemic-SIR, Gossip, Gossip Enforced Ending, Geographic Forwarding Geocast
– Evaluates Range, Latency, Hop Count, Coverage, Neighbors, and Remaining Power
– Emulates an Operations Center User Interface
CPN Methodology
• Small World Networks– Data can be transferred from any node in the network to any
other node in the network in 6 hops or less.
Small World Networks, Courtesy George Washington University
School of Engineering and Applied Science
CPN Analysis
Conclusions:• Optimal Delivery Rate is achieved with Data Buffer Size <300 packets• Upcoming Ad-Hoc Communications Model uses 276 as buffer size
Delivery Rate vs Number of Packets
0.00%
20.00%
40.00%
60.00%
80.00%
100.00%
120.00%
0 150 300 450 600 750 900 1050 1200 1350 1500
Data Buffer Size (# of Packets)
Del
iver
y R
ate
(%)
Communications Networks
• Dedicated/ Fixed – Connection Oriented– Between Tier I Sensors– Between Operations Center and CDC and WHO Headquarters– Between Operations Center and Emergency Response / HAZMAT
Elements– Between Emergency Response Elements– Use Current Technology: Land-line Phone, Trunk Mobile Radio,
WiMax, Fixed LOS, etc.
• Ad Hoc - Connectionless– Between Tier II Sensors and Operations Center/Tier I– Between Tier III Sensors when deployed– Simple Multicast Protocol:
Sense Time Window Communications Time Window
Packet Burst
DestNode ID
SourceNode ID
Detect LongitudeDetect Latitude
A Single Packet: 16 Byte Message
Checksum
Length
64 bits 64 bits
Sensor A and B communicate when in range of each other.
Time required to transmit exactly one packet
… (repeats)time
Powerused
A B
Photos Courtesy KEMRON and Rescue Response Gear
ErrorFlags
ThreatType
ThreatPPM
Ad-Hoc Communications Model Demo
Systems Node Communications Descriptions (SV-2)
Communications Networks (cont.)
Analysis: Range vs. Latency
Communications Range Analysis:Range vs. Latency
0
300
600
900
1200
1500
1800
0 100 200 300 400 500
Communications Range (meters)
La
ten
cy
(s
ec
on
ds
)
Min
Max
Log. (Max)
Conclusions:• Optimal Communications Range: 250m+• Optimal Sensor Range 150m+• Latency can be reduced to under 5 minutes
Optimized Sensor Range Analysis:Range vs. Latency
0
300
600
900
1200
1500
0 100 200 300 400 500
Sensor Range (meters)
La
ten
cy
(s
ec
on
ds
)
Min
Max
Analysis: Hop Count
Optimized Sensor Range Analysis:Range vs. Hop Count
0
1
2
3
4
5
6
7
8
0 100 200 300 400 500
Sensor Range (meters)
Ho
p C
ou
nt
Re
qu
ire
d t
o R
ec
eiv
e A
ll B
io
Th
rea
ts
Min
Max
Conclusions:• Hop Count is roughly linear with Latency• Hop count (when optimized) is six degrees of separation or less: “Small World Communication”
Hop Count vs. Latency, Unoptimized
0
10
20
30
40
50
60
70
0 200 400 600 800 1000 1200 1400
Latency (seconds)
Ho
p C
ou
nt (
Qty
)
Analysis: Neighbors and Coverage
Communications Range Analysis:Range vs. Number of Neighbors
0
10
20
30
40
50
60
0 100 200 300 400 500
Communications Range (meters)
Nu
mb
er
of
Ne
igh
bo
rs
Min Max
Conclusions:• Neighbors increases exponentially with communications range• Coverage increases logarithmically with sensor range• With optimal ranges, neighbors will typically be 0-25 (largely disconnected), and coverage 75% or less
Optimized Sense Range AnalysisRange vs. Sensor Coverage
0.00%
10.00%
20.00%
30.00%
40.00%
50.00%
60.00%
70.00%
80.00%
90.00%
100.00%
0 100 200 300 400 500
Sense Range (meters)
To
tal D
istr
ict
Se
ns
or
Co
ve
rag
e (
%)
Coverage %
Epidemic-SI
0
0.2
0.4
0.6
0.8
1
0 60 120 180 240 300 360 420 480 540 600 660 720 780 840 900 960 1020 1080 1140 1200
Latency (s)
Del
iver
y R
ate
(%)
Epidemic-Susceptible Infective
Conclusions:• As previously demonstrated, Epidemic-SI is quite fast and data can be expected to arrive in 5 minutes or less.• 100% Delivery Rate was achieved for all runs
Algorithm Conclusions
Latency by Algorithm
0
5
10
15
20
Epidemic-SIR Epidemic-SI Gossip EnforcedEnding
Geocast Epidemic-SIS Gossip
Algorithm
Late
ncy
(Min
utes
)
Delivery Rate by Algorithm
0
20
40
60
80
100
Epidemic-SI Gossip EnforcedEnding
Geocast Epidemic-SIS Gossip Epidemic-SIR
Algorithm
Deliv
ery
Rate
(%)
Conclusions:• Epidemic-SIR is best for Latency but worst for Delivery Rate• Epidemic-SI is best overall
Power Conservation by Algorithm
Remaining Tier III Sensor Energy
8450
5080
8680
3150
7150 7050
94309150 8980
82508550
9425
9920 99109560 9720 9800 9900
0
2000
4000
6000
8000
10000
Epidemic-SI Epidemic-SIS Epidemic-SIR Gossip Gossip EnforcedEnding
Geocast
Algorithm Type
Rem
ain
ing
En
erg
y (J
ou
les)
Conclusions:• Epidemic-SI, Epidemic-SIR, Gossip Enforced Ending, and Geocast all have >90% Energy Conservation
Optimal Communications Analysis Conclusions
• Parameters are feasible for current biological sensors in development
• Low sensor ranges provided the best geo-location accuracy
• For speedy delivery performance, “Small World Communications” is best
• A fused DHS Operations Center result is reasonable in under 5 minutes after biological agent detection.
Parameter Optimal Value
Algorithm Epidemic-SI
Latency < 5 Minutes
Communications Range
250+ meters
Sensor Range 150+ meters
Hop Count < 6
Coverage < 75%
Neighbors 0-25, includes disconnection
Delivery Rate > 98%
Data Buffer Size < 300 Packets
Final Thoughts
• Biological Sensor Fusion– Lack of swift response to biological
attack/outbreak scenarios could produce economic loss estimated conservatively at $750M+ and 35+ deaths
– Our system could provide full response within 24-36 hours, preventing deaths and significantly lowering cost for vaccinations, cleanup, and decontamination
– Fusion of data is necessary for responders to target a biological threat real-time
– System demonstrates the use of ad-hoc and mobile ad-hoc communications in a real-world scenario, and is a high interest research area in DoD and DHS
Chicago IllinoisPhoto Courtesy of Destination 360, 2008
Future Work
• Prevention and Treatment– Vaccination Distribution Scenarios– Counter-proliferation Options– Isolation and Treatment Options– Emergency Response Training
• Sensor Research/ Design– Deployment Scenarios– Advanced Technologies:
• UV Fluorescence, Laser-Induced fluorescence, isothermal arrays, genetic classification, electromagnetic spectroscopy, and microfluidics
• Additional Modeling– Biological agent dispersal/ movement– Local sensor processing and data fusion algorithms– Fusion of hospital/medical practitioner data with sensor data– Buffer Size, Cache, Anti-Entropy analyses– Modification of model for other types of EW, ISR or CBRNE sensors– Military Applications
Future Sensors, Pacific
Northwest National
Laboratory
Acknowledgments
• GMU SEOR Homeland Security and Military Transformation Lab– Dr. K.C. Chang
• GMU Faculty and Staff– Dr. Kathryn Blackmond Laskey– Dr. Abbas Zaidi– Others
• City of Chicago– Mr. Earl Zuelke
• U.S. Genomics– Mr. David Hoey
• Cornell University– Dr. Paul Chew
Addressing Risk
• C1: Actual Hours exceed Plan– Mitigation Strategy: (a) Track EVM Weekly to ensure
work is appropriately applied to WBS Tasks; (b) Determine value added tasks on critical path; (c) Discuss analyses tradeoffs with program sponsor
– Risk Closure: End of Performance (EOP)• S1: Actual hours don’t match planned
requirements in WBS categories– Mitigation Strategy: (a) Realign tasks for best
utilization of manhours; (b) Frequent coordination with program sponsor on project scope
– Risk Closure: EOP• P1: Modeling efforts do not support analyses
needed– Mitigation Strategy: (a) Update models with
parameters as needed to support robust analyses; (b) Scope analysis and results with program sponsor
– Risk Closure: EOP• P2: Current Sensor Technology does not support
fast data collection and fusion at an Operations Center– Mitigation Strategy: (a) Coordinate sensor
characteristics with current vendors; (b) Investigate future sensor development plans; (c) Address sensor limitations through modified communications strategies
– Risk Closure: EOP
• P3: Communications Method chosen fails to be adequate for biological threat response– Mitigation Strategy: (a) Alter communications
methodology for Tier I/II/III as needed; (b) Design communications parameters to produce best response times
– Risk Closure: EOP
C1
S1P1
P2
P3
Research Conducted
• DoD and DHS Requests for Proposal (RFPs) on Future Biosensors– Feb 2006: DARPA Biological Warfare Defense Project, $750M+ FY08-FY11– Apr 2004: HSARPA Bioagent Autonomous Networked Detectors (BAND), Rapid
Automated Biological Identification System (RABIS), $48M 18mo periods of performance
• Researched Future Biosensor Development– Johns Hopkins University's
Applied Physics Laboratory of Laurel, MD
– Ionian Technologies, Inc. of Upland, CA
– Goodrich Corporation of Danbury, CT
– Battelle Memorial Institute of Aberdeen, MD
– Physical Sciences, Inc. of Andover, MA
– Research Triangle Institute of Research Triangle Park, NC
– Northrop Grumman Systems Corporation of Linthicum, MD
– MicroFluidic Systems, Inc. of Pleasanton, CA
– Science Applications International, Inc. of San Diego, CA
– U.S. Genomics, Inc. of Woburn, MA
– IQuum, Inc. of Allston, MA – Nanolytics, Inc. of Raleigh, NC – Sarnoff Corporation of Princeton,
NJ– Brimrose Corporation of
Baltimore, MD
Architecture Products
• All Views– AV-1
• Operational Views– OV-1– OV-2– OV-3– OV-5: Node Tree & IDEF0– OV-6c
• System Views– SV-1– SV-2– SV-3– SV-4– SV-5– SV-6– SV-7
System Parameters
• Communications Parameters– Algorithm Used– Range– Burst Time– Tx/Rx Power– Reliability– Data Buffering– Fusion Cordon
• Sensor Parameters– Quantity– Sensor Lat/Long– Sensor Movement– Range– Sensitivity– Specificity– False Positive Rate– Sense Time– Sense Power– Coverage
• Analysis Results– Latency– Delivery Rate– Hop Count– Coverage– Remaining
Power– Neighbors– Algorithm Type
Epidemic-SIS
0
0.2
0.4
0.6
0.8
1
0 60 120 180 240 300 360 420 480 540 600 660 720 780 840 900 960 1020 1080 1140 1200
Latency (s)
Del
iver
y R
ate
(%)
Epidemic-Susceptible Infective Susceptible
Conclusions:• Epidemic-SIS is slower, with Latency results in 1-15 minutes• Two runs only had 90% Delivery Rate, likely due to the delay in becoming susceptible again, although it can be inferred that this data could eventually arrive as long as sensor power remains.
Epidemic-SIR
0
0.2
0.4
0.6
0.8
1
0 60 120 180 240 300 360 420 480 540 600 660 720 780 840 900 960 1020 1080 1140 1200
Latency (s)
Del
iver
y R
ate
(%)
Epidemic-Susceptible Infective Removed
Conclusions:• Epidemic-SIR is the fastest of all algorithms, with Latencies of 3 minutes or less• It is also the worst for Delivery Rate, with 13% of data on average not received
Gossip
0
0.2
0.4
0.6
0.8
1
0 60 120 180 240 300 360 420 480 540 600 660 720 780 840 900 960 1020 1080 1140 1200
Latency (s)
Del
iver
y R
ate
(%)
Gossip
Conclusions:• Gossip is the slowest of all algorithms, with Latencies as high as 20 minutes• One run only achieved 90% Delivery Rate, although it can be inferred that this data could eventually arrive as long as sensor power remains.
Gossip Enforced Ending
0
0.2
0.4
0.6
0.8
1
0 60 120 180 240 300 360 420 480 540 600 660 720 780 840 900 960 1020 1080 1140 1200
Latency (s)
Del
iver
y R
ate
(%)
Gossip Enforced Ending
Conclusions:• Gossip Enforced Ending speeds up latency from regular gossip, but still is only moderately fast, with Latencies in the 3-8 minute range.• One run only achieved 90% Delivery Rate• 100% delivery is not ensured with this algorithm.
Geographic Forwarding Geocast
0
0.2
0.4
0.6
0.8
1
0 60 120 180 240 300 360 420 480 540 600 660 720 780 840 900 960 1020 1080 1140 1200
Latency (s)
Del
iver
y R
ate
(%)
Geographic Forwarding Geocast
•Conclusions:• Geographic Forwarding is slightly faster that Gossip Enforced Ending but not as good as Epidemic-SI• Latencies are typically 3-9 minutes• One run only achieved 90% Delivery Rate• 100% delivery is not ensured with this algorithm.
Conclusions:• At low range, remaining power has a wide variance. This is due mainly to many communications hops and sensing periods, which has a large impact on power.• Low range yielded cases with still very good power conservation in the network.• Communications Ranges beyond 250m+ have little and even sometimes a detrimental effect on power conservation.• With optimal communications, Sensor Range has a slight impact on remaining power, only at ranges <100meters.
Analysis: Power Remaining
Optimized Sensor Range Analysis:Range vs. Tier III Remaining Power
4
5
6
7
8
9
10
0 100 200 300 400 500
Sensor Range (meters)
Re
ma
inin
g P
ow
er
(kJ
ou
les
)
Min
Max
Communications Range Analysis:Range vs. Tier III Remaining Power
0
1
2
3
4
5
6
7
8
9
10
0 100 200 300 400 500
Communications Range (meters)
Re
ma
inin
g P
ow
er
(kJ
ou
les
)
Min
Max
System Performance Parameters
Sensor Design Parameter Optimum Value Communications Design Parameter
Optimum Value
Tier I Sensor Quantity 6 Communications Algorithm Epidemic-SI
Tier II Sensor Quantity 120 Communications Range 250m+
Tier III Sensor Quantity 150 Communication Time Window 20 Sec
Sensor Range 150m+ Communications Burst Time (per packet)
5mS
Sensitivity 99.9%+ Minimum Time Step 1mS
Sensitivity Error <0.5% Tier I Transmission Power 50W
Specificity 99.9999999%+ Tier II Transmission Power 5W
False Positive Rate <1E-7% Tier III Transmission Power 10mW
Sense Time <4 hours Receive Energy 10J
Sense Data Updates Every 120 Secs Multi-path Loss 2.5%
Tier III Sense Energy 50J Latency <5 mins
Tier III Total Available Energy 10000J Hop Count <6
Sensor Coverage 35%+ Delivery Rate 100%
Neighbor Quantity 0-25, Includes Disconnection Data Buffer Size <300