Data Collection with In-network Fault Detection Based on Spatial Correlation

Environmental sensing exposes sensor nodes to environmental stresses that can lead to various kinds of sampling failure. Recognising such faults in the network can improve data reliability therefore making sensor networks suitable candidate for critical monitoring applications. We develop a technique that builds a spatial model of a sensor network and its observations, and show how this can be updated in-network to provide outlier detection even for non-stationary time series. The solution does not require local storage of learning data or any centralised control. The method is evaluated by both real world implementation and simulation, and the results are promising.