Intelligence Surveillance and Reconnaissance System for California Wildfire Detection

1
Intelligence Surveillance and Reconnaissance
System for California Wildfire Detection
Team ISR Firefighting Team members Shashank
Tamaskar Nadir Bagaveyev Evan Helmeid Tiffany
Allmandinger

• Presented by-
• Shashank Tamaskar
• Purdue University
• stamaska_at_purdue.edu

2
Overview

1. Definition
2. Abstraction
3. Modeling and Implementation
4. Future work

3
Definition

• Need In 2008, Arsonist fires burned down 25,000
  acres of forest land resulting in 24 million
  dollar damage to property
• Objective To understand and analyze the problems
  associated with the wildfire prevention and
  management system and to suggest improvements to
  enable faster fire detection in the region
• SoS traits Heterogeneous geographically
  separated agents (Watchtowers, UAVs, arsonists,
  other human agents) with different degree of
  autonomy
• Status Quo Current system consists of
  watchtowers and the reliance on civilian reports.
  Intelligence of multiple fires and fire state
  dependent on ground crews or manned aircraft
  scouting situation. Manned airplanes limited in
  allowed exposure to fire conditions. No night
  flying allowed.
• Operational context To limit the scope of the
  project we have concentrated on interaction
  between the resource and operational alpha level
  entities. The following figure shows our area

4
Definition
5
Abstraction
-Framing key descriptors and their evolution
Paper Model
6
Abstraction
UAV Path Calculations Different operating
scenarios Zigzag model AOI divided into
sectors Coverage due to watchtowers
ignored Waypoints predefined ABM Waypoints
dynamically added depending upon coverage UAVs
avoid watchtowers Optimal Path Generation Optimal
path generation to maximize the coverage
7
Abstraction Zigzag model
8
Abstraction ABM
9
Implementation

• Platform STK is used for calculation of
  positions of mobile agents while Matlab is used
  to implement path algorithms and calculate
  coverage
• Object orientation programming is used to rapidly
  develop large code
• (gt1000 loc) also the modular architecture of the
  code helps us keep the effective complexity of
  the code low
• Metrics Four metrics for system performance
• Coverage
• Cost
• Detection Time
• Response time

10
Implementation

• Problems addressed
• Coverage Problem How to efficiently provide
  coverage to a area given a set of assets
• Detection Problem How to improve the fire
  detection time
• Random fires
• Arsonist fires
• Arsonist tracking Track the arsonist after the
  fire is detected

11
Implementation
Demonstration of the model
12
Implementation
Simulation Results Verification/Validation

• For a constant field of view
• 1 UAV provides worst coverage
• 2 to 5 UAVs do not present significant coverage
  differences

• Coverage metric is directly related to the
  detection time over all simulations

13
Implementation
Simulation Results

• 1 UAV has the worst detection time
• 10o, 15o FOV cause significant increase in
  detection time over 20o, 45o
• 20o FOV provides the best coverage for the cost
• Cost is directly related to the FOV
• Small FOV yields a highly unstable system and
  requires many more simulations to determine
  trends
• The larger FOV follows the expected trend more
  UAVs ? faster detection

14
Implementation
Simulation Results

• Determine effectiveness of watchtowers and impact
  on UAV necessity
• UAVs detected the majority of the fires
• Provide significant increase in system
  performance over the current state
• As the number of simulations were increased the
  fraction of fires detected by the watchtowers
  became even less
• Watchtowers are good for random fires but UAVs
  are good for arsonist fires. UAVs also allow for
  arsonist tracking

15
Implementation
16
Implementation
Simulation Results

• 10o and 15o FOV do not provide low enough
  detection time
• 20o and 45o are the most effective
• 20o is the most cost effective
• Best performance for the money
• 45o does not provide enough benefit increase to
  justify increased cost over the 20o FOV

17
Implementation
Simulation Results Arsonist detection

• Only 19 out of 132 simulations resulted in
  arsonist detection
• In most cases fires were detected late after they
  started so arsonists had sufficient time to flee
  the scene
• Probability of arsonist detection increases with
  increase in number of UAVs, Speed, FOV and
  altitude
• Arsonist detection by Humans was reported too late

18
Conclusions

• Simulation
• Effective and valid Simulations show a good
  correlation with paper model
• Metric is accurate Found a good correlation
  between detection time and coverage metric
• More iterations and simulations are necessary to
  draw proper conclusions
• Generated a model which can be applied to other
  ISR problems
• Conclusions
• UAVs with greater FOV and Altitude can
  significantly improve the detection time
• More UAVs provide better coverage, but do not
  necessarily provide significant benefits
• Arsonist detection may better suited with a fleet
  mix of UAVs. Slow UAVs for fire detection and
  Fast for Arsonist detection
• Watchtowers are not well suited for detection of
  arsonist fires
• A SoS approach is beneficial in analyzing the
  options for improving the current system, but it
  may not be feasible to implement the SoS

19
Future Work

• Implement UAV-avoidance algorithm
• Do not revisit areas that were just scanned
• Limit conflict between UAVs
• Consider refueling time
• Create a detailed cost model
• Determine camera/sensor array to use
• Determine optimal UAV for given parameters
• Pool together the lessons learned by various
  teams and develop a general purpose tool for ISR
  applications which can be used for research at
  Purdue

20
System-of-Systems Laboratory Aeronautics
Applications Director Prof. Dan DeLaurentis
(ddelaure_at_purdue.edu)
21
Thank you for your consideration!