Fuzzy Based Landslide Prediction using Wireless Sensor Networks

1
Fuzzy Based Landslide Prediction using Wireless
Sensor Networks

• Mohamed Shahim P.I.
• Guide Prof. U.B. Desai
• Spann-Lab, IIT Bombay

2
Introduction

• Methods for modeling Uncertainties
• - Probability
• - Fuzzy Logic
• - Fuzzy Inference systems
• Advantages of Fuzzy
• - Easy to model
• - More easy to change the behavior of the
  system, by changing the rules.
• - Less computations compared to hypothesis
• testing methods

3
Fuzzy logic

• Fuzzy Sets
• Fuzzy operaters.
• Eg AND -
• Fuzzy if-then rules
• if (i/p1 is X) then (o/p1 is Z)

antecedent
consequent
4
Fuzzy Inference System
O/P
Rule Evaluator
Aggregators
I/P

• Fuzzification Real values of the input into
  membership values in appropriate Linguistic
  variables.
• Rule base evaluation Each of the rules is
  evaluated to get an output modified fuzzy set
• Aggregation All the output fuzzy sets are
  combined to form a single set using max rule
• Defuzzification Combined output fuzzy set is
  converted to a number. E.g.. Centroid calculation

Fuzzifier
De-Fuzzifier
5
FIS information flow
6
Fuzzy system model

• In our case, input is strain values from the
  sensors
• It is divided into 3 linguistic variables as
  shown below
• Output Hazard is also divided into 3 variables

7
Input membership functions
8
Stress-Strain Characteristics
9
Output membership functions
10
Rule base

• A total of 81 rules for 4 node setup

11
Simulations

• Setup consists of 4 nodes and 1 Base Station
  (BS).
• At the BS, FIS calculates output hazard level.
• Simulated using MATLAB Fuzzy Toolbox

12
ROC

• ROC gives the noise performance of the system
• The data generated using the VMGP model is used
  as input. Output hazard varies from 0 to 10. It
  is converted to 2 outputs, Landslide and No
  Landslide, using a simple threshold.
• The threshold is varied to plot the entire ROC.

13
(Input Strain Noise) vs Time
Output Hazard vs Time
14
ROC comparing Fuzzy CVBD for SNRs 10 dB, 20
dB, 30 dB.
15

• Previous system uses space correlation, as can be
  seen in the rules.
• Use of time correlation also helps in removing
  wrong predictions.
• One simple method is to remove outliers in output
  Hazard by averaging.

16
ROC after Averaging the o/p Hazard
17
Issues

• The region will be already in a stressed state. A
  strain gauge can only measure further change.
  Difficult to calculate whether rock has reached
  70 of breaking stress.
• Heterogeneity of rocks in the Landslide prone
  region.
• - This makes fixing of the threshold a
  difficult task.
• Performance variation of strain gauges with
  respect to temperature.
• - Use of Temperature compensation circuits
  for interfacing
• Effect of Orientation of strain gauge on measured
  stress.
• - Strain Gauge Rosette measures greatest
  strain at a Point.
• Noise in Measured Strain

18
Future Work

• Use of FLAC3d software for simulating a slope
  failure
• Extending a single cluster to a bigger network to
  form a Distributed sensor network
• Safety Factor
• Distance between motes can be used for
  detection movement in a slope. Helps in detecting
  rock movement
• Distance can be measured using Received power
  (RSSI).

19
References

• Wang Y.,T.Y. Yu and D. Andra, Tornado
  detection using a neuro-fuzzy method,(
  Albuquerque), 32nd Conference on Radar
  Meteorology, NM. American Meteorology Society,
  October 2005
• T. Ross, Fuzzy logic with Engineering
  Applications. Hightstown, NJ McGraw-Hill, 1995
• The Math works Inc. Fuzzy Logic Toolbox Users
  Guide
• H.B. Wang and R. Xu, Slope stability evaluation
  using back propagation neural networks,
  Engineering Geology, vol.80, pp. 302-315, June
  2005.
• FLAC3d Home Page, 2002. http//www.itascacg.com/
  flac3d.html