Distributed Fault Detection of Wireless Sensor Networks

Chen, Kher, Somani

sensor networks fault management detection

@inproceedings{chen:DIWANS-2006,
  title={Distributed Fault Detection of Wireless Sensor Networks},
  author={Chen, J. and Kher, S. and Somani, A.},
  booktitle={Workshop on Dependability Issues in
             Wireless Ad Hoc Networks and Sensor Networks},
  pages={65--72},
  year={2006},
  organization={{ACM}}
}

Testing usually done via built-in self test or signature verification subsystem

• Sometimes combined with built-in self repair, as in DRAM

If neighbors are testing neighbors, node degree must be reasonable to ensure useful diagnoses

• Majority vote among neighbors

Watchdog processors monitoring control flow, memory access behavior

Identiying insecure nodes

Pushing reports on state change

A sensor is likely to be faulty if removing its data improves the consistency of the remaining results

• Similarly, apply maximum likelihood or Bayesian to determine if combined results are improbable
• Even measurements are likely to be spatially correlated
• Noise and sensor faults are likely to be uncorrelated
• Simply comparing to median value in area is easy and effective, but only if it is likely that less than half the nodes are in error

Voting requires each sensor has at least 3 neighbors

Various types of error

• Random individual failure vs widespread area failure
• Systematic calibration error vs random noise vs complete malfunction