Underground Structure Monitoring with Wireless Sensor Networks

1
Underground Structure Monitoring with Wireless
Sensor Networks
Mo Li, Yunhao Liu Hong-Kong University of Science
and Technology limo,liu_at_cse.ust.hk

• Date 06th Dec. 2007 Presenter KM Chen

2
Outline

• Motivation
• Overview of Structure-Aware Self-Adaptive sensor
  system (SASA)
• Detecting and locating the collapse hole
• Accident reporting
• Displaced node detection and reconfiguration
• Hardware
• Application Scenario
• Experiment and Performance
• Simulation
• Related Work
• Conclusion

3
Motivation

• Over the last decade, collapses account for more
  than 50 of fatalities in U.S. in coal mine
• The unstable nature of geological construction in
  coal mines makes underground tunnels prone to
  structure changes.
• Environment monitoring in underground tunnels had
  been a crucial task to ensure safe working
  conditions in coal mines.
• There is a need to develop a wireless sensor
  network system to quickly detect the collapse
  hole regions and accurately provide location
  references to evacuate workers from the dangerous
  zone.

4
Outline

• Motivation
• Overview of Structure-Aware Self-Adaptive sensor
  system (SASA)
• Detecting and locating the collapse hole
• Accident reporting
• Displaced node detection and reconfiguration
• Hardware
• Application Scenario
• Experiment and Performance
• Simulation
• Related Work
• Conclusion

5
Design of SASA (1/3)

• a) Stationary sensor nodes are deployed on the
  walls and ceiling of tunnels to form a mesh
  network.

6
Design of SASA (2/3)

• b) - Unfold 2 walls of the tunnel and builds a
  2-D representation of the 3-D deployment on the
  inner surface of the tunnel.
• - Nodes are configured with 2-D coordinates on
  the unfolded 2-D surface then transformed into
  the 3-D corresponding locations with the
  knowledge of longitudinal section.

7
Design of SASA (3/3)

• c) - The distance between 2 nodes in the 3-D real
  environment is less than or equal to the distance
  between the pair in the unfolded 2-D view.
• - The real connectivity of the sensor network is
  no less than shown in the 2-D representation.

8
Definitions and Theorem

• Edge node A node defines itself as an edge node
  if the 2 adjacent neighbor nodes are detected
  lost during a time period.
• Hole Polygon The largest polygon outlined by the
  collapsed sensor nodes with every edge ending at
  2 adjacent nodes.
• Theorem The convex hull of edge nodes in SASA
  encloses the hole polygon.

Convex Hull
Hole Polygon
Edge node (2 adjacent neighbors are detected lost)
9
Detecting and locating the collapse hole

• Goal Let sensor nodes to maintain a set of their
  neighbors. When nodes detect a loss of neighbors,
  a hole is detected.

Collapse hole
10
Detecting and locating the collapse hole

• Question How to maintain a neighboring set?

• Node Beaconing Mechanism
• Each node maintains a neighboring nodes list in
  memory.
• Each node periodically broadcasts beacon messages
  that include its location.
• Upon receiving a beacon message, the node updates
  the corresponding entry.
• If a node fails to update an entry in a fixed
  time interval, then it represents the loss of the
  neighbor

11
Detecting and locating the collapse hole

• Problem
• It was observed that the neighbor set of a node
  is highly unstable, even if all the nodes work
  normally.
• Solution
• SASA deploys sensor nodes in a cellular hexagonal
  placement such that the node distribution is
  uniform.
• Every pair of adjacent nodes are separated by the
  same interval .
• Each node is limited to maintain a neighbor set
  to the 6 adjacent nodes.

12
Detecting and locating the collapse hole

• How to define a hole detection?
• Only failure of at least 2 adjacent nodes are
  necessary to define an edge node. Nevertheless,
  if 2 adjacent nodes fail simultaneously, a hole
  is detected
• What about single node failure?
• Unfortunately a small hole affecting only 1
  sensor node can not be detected.

13
Accident reporting

• Goal When edge nodes detect a hole, they report
  to the sink with the locations so that the hole
  can be illustrated by calculating the convex
  hull.
• Problem
• Create traffic peak and increase collision
  domain.
• Solution Randomized Forward Latency and Data
  Aggregation
• Insert a flag into the beacon messages, which
  indicates whether the beaconing node is an edge
  node.
• Upon receiving other edge nodes beacon message,
  an edge node records them locally.
• When this edge node sends out its report message,
  it aggregates all the recorded locations of its
  nearby edge nodes.
• If an edge node receives a report message
  containing its own location, it simply forward
  this message instead of creating a new one.
• The sink will send out reply to limit the number
  of retransmission.

14
Displaced node detection and reconfiguration

• Goal Rapidly detect displaced nodes and
  reconfigure with correct locations in order to
  maintain system validity.
• Centralized approach
• When the sink receives report messages with the
  edge nodes locations and approximate the hole
  region, it broadcasts the convex hull area.
• Every node within the convex hull will start
  detecting its surroundings and check its location
  from beacon message.
• If the 2 locations differ beyond some threshold,
  then it knows its being displaced.

15
Displaced node detection and reconfiguration

• Distributed approach
• There are 3 types of edge nodes
• Edge nodes that lose neighbors but themselves do
  not move
• Since their locations are correct, they dont
  need to be reconfigured
• Edge nodes that fall into an area where no normal
  node exists
• They have no impact on normal nodes, they do not
  need to be reconfigured either.
• Edge nodes that fall into other normal node range
• Stop beaconing . This operation will lead the
  neighboring displaced nodes to become edge nodes,
  if they are not yet.

16
Displaced node detection and reconfiguration

• Centralized approach
• Advantage Short latency when the hole is closer
  to the sink
• Disadvantage May suffer long latency and low
  accuracy due to high link loss rate in coal mine,
  especially when a collapse area in a long tunnel
  is far from the sink.
• Distributed approach
• Advantage and disadvantage Independent of the
  distance to the sink.
• In summary
• Combining both algorithms provides efficient and
  reliable for various situations
• Turn them off or reconfigure their locations to
  conform to their new positions.

17
Displaced node detection and reconfiguration

• Location Calculation
• Suppose A and B drop into a new area surrounded
  by 3 resident nodes.
• When A first detect the surrounding 4 nodes, it
  calculates a new location as (32.5, 19.25) and
  replaces the original locations.
• When B detects its surroundings, it utilizes the
  new location of A and calculate a new location as
  (15.63, 11.56).
• When A iteratively calculates its new location,
  it will get a more accurate result of (11.41,
  7.14)

18
Displaced node detection and reconfiguration
(10, 17)
A (100, 100)
A (32.5, 29.25)
A (11.41, 7.14)
(20, 0)
(0, 0)
B (100, 120)
B (15.63, 11.56)
19
Outline

• Motivation
• Overview of Structure-Aware Self-Adaptive sensor
  system (SASA)
• Detecting and locating the collapse hole
• Accident reporting
• Displaced node detection and reconfiguration
• Hardware
• Application Scenario
• Experiment and Performance
• Simulation
• Related Work
• Conclusion

20
Hardware

• Mica2 platform developed at UC Berkeley
• The MPR400 radio board employed has a 7.3 MHz
  microprocessor.
• 128K bytes of program flash memory.
• 512K bytes of measurement flash memory.
• 868/916 MHz tunable chipcon CC1000 multi-channel
  transceiver with a 38.4 kbps transmitting rate is
  employed for wireless communication with a 500
  foot outdoor range

21
Outline

• Motivation
• Overview of Structure-Aware Self-Adaptive sensor
  system (SASA)
• Detecting and locating the collapse hole
• Accident reporting
• Displaced node detection and reconfiguration
• Hardware
• Application Scenario
• Experiment and Performance
• Simulation
• Related Work
• Conclusion

22
Application Scenario

• Cooperate with S.H. Coal Corporation and selected
  the D.L. coal mine as the experimental
  environment.
• D.L. coal mine is the is one of the mist
  automated coal mines, yielding the second largest
  production of coal worldwide.
• Slightly sloped 14-kilometer long main tunnel
  from the entrance above the ground surface and
  goes 200 meter deep underground.

23
Application Scenario

• Requirements for SASA implementation in D.L. coal
  mine
• Remote management Remotely maintain and manage
  the entire monitoring system, efficient and
  robust communications and routing mechanisms are
  required under all conditions.
• In-Situ interactions Besides stationary sensors
  deployed on the walls, poles and floors, miners
  carry mobiles sensors providing real-time
  geographical references.
• Awareness of structure variations Using node
  collaborating mechanism for collapse detection.
• Maintenance of system validity Maintaining the
  validity of the monitoring system in extreme
  situation.

24
Outline

• Motivation
• Overview of Structure-Aware Self-Adaptive sensor
  system (SASA)
• Detecting and locating the collapse hole
• Accident reporting
• Displaced node detection and reconfiguration
• Hardware
• Application Scenario
• Experiment and Performance
• Simulation
• Related Work
• Conclusion

25
Experiment and Performance

• A prototype system with 27 Mica2 motes is
  implemented in the D.L. coal mine.
• It is distributed on a tunnel wall about 12
  meters wide and 5 meters high.
• Nodes are pre-configured with their location
  coordinates.
• Nodes are placed in hexagonal mesh regulation

26
Experiment and Performance

• Hole detection percentage A hole is counted as
  undetected if less than 3 nodes reports are
  received by the sink.
• Hole detection error The error in distance
  between the real and detected position of the
  hole region.
• Reconfiguration error (2-D or 3-D) Localization
  error in the reconfiguration process.

27
Experiment and Performance

• Over 80 of the detected holes are located within
  1 meter from its real position and 99 are less
  than 2 meters.

• All the 2-D reconfiguration errors and over 80
  of the 3-D reconfiguration errors are below 3
  meters

28
Experiment and Performance

• Detection latency Time from when the hole
  emerges until it is detected
• Turn-off latency The latency when the displaced
  nodes were turned off.
• Reconfig latency Latency when reconfiguring the
  displaced nodes according to the normal nodes
  surrounding them.
• Short beacon interval leads to short latency.
  However, frequent beaconing brings large
  overhead, heavy collisions and increased packet
  loss

29
Experiment and Performance

• The packet loss rate rapidly drops as the beacon
  interval increases while under short beacon
  intervals (0.8s), then becomes stable around a
  fixed level.
• The loss rate is heightened as the exerted
  traffic overhead increases.

30
Experiment and Performance

• Observations
• We can carefully select a proper beacon interval
  for a specific application workload to balance
  communication quality and the processing latency
• Shorter beacon interval to reduce the processing
  latency if the application workload is light.
• Longer beacon interval to reduce the packet loss
  rate if the application workload is heavy.

31
Outline

• Motivation
• Overview of Structure-Aware Self-Adaptive sensor
  system (SASA)
• Detecting and locating the collapse hole
• Accident reporting
• Displaced node detection and reconfiguration
• Hardware
• Application Scenario
• Experiment and Performance
• Simulation
• Related Work
• Conclusion

32
Simulation

• 2000 nodes were simulated on a 1000m x 20m plane.
• Nodes were placed in a hexagonal mesh regulation
  with 3 meter interval between each node.
• A transmitting rate of 16 packet/s is used in the
  simulation for the nodes communication channels.

33
Simulation

• When the hole size increases, the outline of the
  edge nodes becomes tighter therefore the
  precision is dramatically increased.

• Detection error is stable as slightly decreases
  as the hole size increases.
• Larger hole includes more edge nodes, giving a
  more accurate outline of the hole region.

34
Simulation

• When the hole is close to the sink, the
  centralized algorithm benefits from rapid
  information collection and reaction from the
  sink.
• When the hole is far away from the sink,
  centralized algorithm suffers from the round-trip
  time from the sink. The distributed algorithm is
  not affected.

• As the packet loss rate between any 2
  communicating nodes, and the random node failure
  rate increase, the misreport ratio also
  increases.
• Need to decrease the beacon frequency in order to
  preserve a better communication channel.

35
Outline

• Motivation
• Overview of Structure-Aware Self-Adaptive sensor
  system (SASA)
• Detecting and locating the collapse hole
• Accident reporting
• Displaced node detection and reconfiguration
• Hardware
• Application Scenario
• Experiment and Performance
• Simulation
• Related Work
• Conclusion

36
Related Work

• Wireless sensor networks for habitat monitoring
  A. Mainwaring, J. Polastre, R. Szewczyk, D.
  Culler and J. Anderson
• A Wireless Sensor Network for Structural
  Monitoring N. Xu, S. Rangwala, K. K.
  Chintalapudi, D. Ganesan, A. Broad
• Hole problem in wireless sensor network N. Amed,
  S. S. Kanhere and S. Jha
• Coverage hole
• Routing hole
• Jamming hole
• Sink/black/worm holes

37
Outline

• Motivation
• Overview of Structure-Aware Self-Adaptive sensor
  system (SASA)
• Detecting and locating the collapse hole
• Accident reporting
• Displaced node detection and reconfiguration
• Hardware
• Application Scenario
• Experiment and Performance
• Simulation
• Related Work
• Conclusion

38
Summary and Future Work

• By regulating the mesh sensor network deployment
  and formulating a collaborative mechanism based
  on the regular beacon strategy, SASA is able to
  rapidly detect structural variations caused by
  underground collapses
• The collapse holes can be located and outlined,
  and the detection accuracy is bounded. We also
  provide a set of mechanisms to discover the
  relocated sensor nodes in the hole region.
• How to organize mobile nodes to form efficient
  collaborative groups is a challenging issue.
• Single hole detection.