Online data fault detection in wireless sensor networks

The critical applications of wireless sensor networks, the increased data faults and their impact on decision making reveal the importance of adopting online techniques for data fault detection and diagnosis. Keeping in mind the hardware limitations of sensors, this work focuses on complementary signal processing techniques (temporal, spatial correlation and self organizing map) in order to cover several types of data faults, reduce the misdetection rate and also isolate faults when possible by specifying the defaulting sensors. The methods applied to a real database show that 31.6% of data are faulty by applying SOM3D in conjunction with the spatial correlation. The combination of the above technique in addition to the temporal correlation reduces the misdetection by increasing the detection percentage by 17.6%. SOM3D model also helped identifying the least trustful sensors among the network sensors, this can be helpful when reconciling errors.