Neural fault isolator for Wireless Sensor Networks

Wireless sensor networks are emerging as an innovative technology that can help to improve business processes. In such environments malfunctions and break-down states must be efficiently diagnosed to reduce to a minimum the economic losses. In this paper we present a fault isolation approach based on neural networks, which utilizes only a minimum set of information such as the sensor value, node ID and timestamp as inputs. We believe that this information set could be provided by any WSN regardless of its specific implementation. This abstraction makes the fault isolator generically applicable in enterprise business systems. The neural fault isolator was evaluated in a trial with 36 nodes and has proved to be highly efficient in the isolation of failed components.