A fault inference mechanism in sensor networks using Markov Chain

The reliability of communication and sensor devices has been recognized as one of the crucial issues in Wireless Sensor Networks (WSNs). In distributed environments, micro-sensors are subject to high-frequency faults. To provide high stability and availability of large scale sensor networks, we propose a fault inference mechanism based on reverse multicast tree to evaluate sensor nodes' fault probabilities. This mechanism is formulated as maximization- likelihood estimation problem. Due to the characteristics (energy awareness, constraint bandwidth and so on) of wireless sensor networks; it is infeasible for each sensor to announce its working state to a centralized node. Therefore, maximum likelihood estimates of fault parameters depend on unobserved latent variables. Hence, our proposed inference mechanism is abstracted as Nondeterministic Finite Automata (NFA). It adopts iterative computation under Markov Chain to infer the fault probabilities of nodes in reverse multicast tree. Through our theoretical analysis and simulation experiments, we were able to achieve an accuracy of fault inference mechanism that satisfies the necessity of fault detection.