Content Based Pre-diagnosis for Wireless Sensor Networks

The network diagnosis is significant and difficult because the limited resource in WSNs, especially the heavy communication burden for diagnosis. How to reduce the diagnosis overhead is crucial. In this paper, we argue that the sensing data from sensor networks presents another new model called DSD for network diagnosis. We discover that the character of the sensing data reflect the network status in some way, according to considerable experiments in the Green Orbs project. We mine the hints between the sensing data and the failure in the sensor networks, and maintain a failure knowledge library which can update through self-learning. Through this diagnosis mechanism, we deduce the node failures and the network failures without any additional network burden, and the final diagnosis report can give the network administrators good reference for further diagnosis. In this way, we make full use of the sensing data to do the pre-diagnosis according to the characteristics of the sensing data. This approach avoids sending metrics periodically from sensor nodes. Moreover, the failure library can improve the efficiency of diagnosis. We analysis the three months sensing data from the Green Orbs project, analytically and experimentally, we show that the proposed scheme can improve the diagnosis performance with low energy cost.