Improving Data Recovery From Embedded Networked Sensing Systems with Fault Detection and Diagnosis

Опубликовано 6 сентября 2016, 17:00

Sympathy and Confidence are new fault remediation systems that improve the quality and quantity of data recovered from embedded networked sensing (ENS) systems. Unreliable collection systems result in missing data points, and faulty or confusing sensor data further reduce the quantity of usable data recovered from a deployment. Unfortunately faults in ENS systems occur at a higher rate, and are harder to find and fix, compared with conventional systems. I will talk about two methodologies that we have developed to find and fix faults in ENS deployments. Sympathy uses a high-level decision tree to analyze a small set of carefully chosen metrics and to suggest actions users can take to fix data-flow disruptions in the network. Sympathy has been effective for managing network health in numerous real-world deployments since 2006, but its static algorithms proved insufficient for data faults. Our second system, Confidence, combines the successes of Sympathy with dynamic algorithms that more readily adapt to detecting data faults in new and unexpected environments. Confidence builds on three key ideas: 1) A flexible, multidimensional feature space that tends to group nodes and sensors that have similar fault states; 2) a transparent system design that both aids and profits from the user's intuition; and 3) efficient incorporation of user feedback into fault detection and diagnosis algorithms. User feedback, in particular, makes Confidence quickly adapt to new deployment scenarios and previously unobserved types of faults. We have successfully deployed both Sympathy and Confidence alongside many ENS deployments in California and Bangladesh, including a long-term deployment at James Reserve of 130 sensors.