Fault classification and model learning from sensory Readings — Framework for fault tolerance in wireless sensor networks

Primary task of wireless sensor networks is to deliver reliable and accurate information about the phenomena of interest. However, faults are a frequent occurrence and their accumulation affects the quality of service significantly. This leads to a shorter effective lifetime of the network. In this work, we propose a framework for the fault tolerance in sensory readings. The main concept is based on the observation of the pattern that faults leave in data behavior. Based on the duration, continuity and the impact, we propose a complete and consistent classification of faults as they can be observed in sensory readings independently of the underlying cause. Further, we propose that network learns a model of a fault for each faulty node from the past behavior. Each phase of the framework can be implemented with the use of different algorithms appropriate for the task. In this paper we present an instance that relies on neighborhood vote, time series analysis and statistical pattern recognition. Results so far confirm that the scheme works well for dense data-centric wireless sensor networks.