Node-level indicators of soft faults in wireless sensor networks

Faults in wireless sensor nodes tend to be the norm rather than the exception. Our paper addresses the following problems: How can we make sure the sensed data has not been distorted by faults? And when we experience anomalies in the sensed data, how can we distinguish between rare but correctly measured events and incorrect data due to faults? The focus of this paper is specifically on soft faults, because (i) they deteriorate the data quality and (ii) are hard to distinguish from irregular but correct events. Many papers on how to detect soft faults have been proposed, but they mostly rely on assumptions on the network's deployment or the nature of the collected data. Therefore, they are insufficient as they can miss faults or misinterpret correct data. In this paper, our key idea is to augment the existing fault-detection approaches with node-level information as additional input. Our contribution is to show the existence of such node-level information that can improve (i) the detection of soft faults and (ii) the distinction between faults and events. This node-level information - called fault indicators - can then be included in existing fault detection schemes. Based on practical experiments, we show that using such fault indicators indeed improves the detection rate and reduces the risk of missing fault-induced variations of sensor data.