Data-Driven Cross-Layer Fault Management Architecture for Sensor Networks

Abstract

The paper proposes a data-driven cross-layer resilient architecture for sensor networks. The novelty of the approach lies in combining fault detection across data and network layers into a coordinated system health management architecture.The implemented fault detection is entirely data-driven: data are collected exclusively by the functional sensors that are part of the system. Thus, there is no need for additional hardware resources.The data layers considered include the raw sensor data layer, the processed data layer and the data aggregation layer. The proposed cross-layer fault management architecture utilizes a hierarchical health-map structure for fault detection and data aggregation. A practical case study of an underwater sensor network for harbor water flow monitoring application based on the proposed architecture is presented. Synthetic experiments with real data demonstrate the effectiveness of the approach in fault detection and diagnosis. The experiments show that the data-driven cross-layer fault management allows improving the sensor group measurement accuracy by 35% in case of single sensor errors and nearly twofold in case of double sensor errors. The paper also presents examples of system health-map aggregation and fault diagnosis based on faults manifesting at the different layers for real incidents occurring in the field.