Bayesian Selection of Non-Faulty Sensors

The identification of sensors returning unreliable data is an important task when working with sensor networks. The detection of these unreliable sensors while in the field can cue human involvement in repairing problem sensors. This ensures that meaningful data is collected throughout the entire length of a sensor deployment. We present a detection based method of identifying faulty and non-faulty sensors from a given set of sensors that are expected to behave similarly. We use a Bayesian detection approach to select a subset of sensors which give the best probability of being correct given the data. This gives us a model from which we can determine whether sensors´ readings fall out of a reasonable range for the sensor set. We apply our method to simulated data and actual environmental data collected in the forest.