Research On Efficient Methods For Faulty Node Detection In Sensor Networks

In recent years,wireless sensor networks have been successfully used for environmental monitoring,vehicle tracking,etc.Sensor nodes are prone to failure due to its low cost and harsh setting environment.The monitoring capability and data processing accuracy of the entire network is strongly affected by faulty sensor nodes.Consequently,detecting fault nodes quickly and conveniently is necessary to ensure the efficient work of entire sensor networks.It is convenient to detect failure sensor node for it stops collecting and submitting data.While detecting fault sensor node is harder for fault sensor node still submits collected data.The data collected by fault sensor node is worry or with excessive noise.Most of the existing methods for detecting fault sensor nodes use the spatio-temporal correlation of data collected by nodes.However,the spatio-temporal correlation of data collected by nodes is low in bridge and architecture system which made these methods unable to detect accurately.This thesis studies using collected sensor data to detect fault sensor nodes in the bridge or architecture system without adding extra collected data of sensor network.The main goals of this paper can be divided into the following three aspects:1.The fault detection method based on spatio-temporal comparison is no longer applicable for the direct spatio-temporal correlation of data collected by bridge or architecture monitoring sensor network nodes is poor.A fault detection method based on Kalman filter reconstruction error is proposed to solve this problem.The proposed method uses Kalman filtering and system initial state space parameters to process measurement data from a group of sensor nodes.By judging whether the Kalman filtering reconstruction error is greater than a preset threshold,we can conclude whether there are fault nodes in this group.Experiments on measured data of bridge monitoring sensor networks show the applicability of this fault detection method based on Kalman filter reconstruction error.2.Research on fault node location method is based on group testing.Using group testing technology,we design several methods for selecting group testing vectors and apply this method for locating fault individuals according to multiple grouping test results.This method can make the test time much less than the amount of sensors in the network.By combining group testing technology with fault detection method based on Kalman filter reconstruction error,a new fault node location method in sensor network can be realized.3.We also proposed an adaptive Bayesian group testing algorithm,which solved the misjudgment problem in the fault detection method based on Kalman filter reconstruction error.Bayesian theory is introduced into this group testing algorithm and probabilistic reasoning is applied instead of deterministic reasoning.Test vectors are adaptively selected according to the maximum expected value of update gain.While the update of sensor node state probability is obtained according to Bayesian theory.Experiments on measured data of bridge monitoring sensor networks show that the performance of fault node location algorithm based on adaptive Bayesian group testing needs less test time than deterministic reasoning method.Meanwhile,fault node can still be located accurately even if there is misjudgment in group testing.This algorithm is proposed to improve the robustness of fault node location and it has high robustness.