Trust-Enhanced Distributed Kalman Filtering for Sensor Fault Diagnosis in Sensor Networks

Sensor fault diagnosis is a critical issue in Sensor Networks (SNs) since sensor failures could lead to significant errors in data fusion and state estimation. To address this challenge, we propose a trust-enhanced distributed Kalman filter (TeDKF) designed to improve the state estimation performance of SNs under sensor faults. The TeDKF framework incorporates a novel incremental density-based (IDB) clustering mechanism into the distributed diffusion Kalman filter (DDKF) structure, which can support an intermediate-level feature (innovations) exchange and effectively fuses reliable sensor nodes. Unlike conventional clustering schemes, IDB clustering does not rely on majority voting, where more than half of the nodes must be reliable. Instead, it can effectively detect and eliminate faulty sensors even in scenarios where the majority of nodes are compromised. This dynamic clustering builds-up trust by selectively grouping the reliable nodes based on evolving normal system behavior, which is considered as a dynamic trust reference to detect anomalies and isolate faulty sensors irrespective of majority voting. The experimental results show that TeDKF significantly reduces estimation errors and enhances fault tolerance compared to the traditional Kalman filtering technique. It can handle different sensor faults, like bias, drift, noise, and stuck faults, especially in scenarios where most nodes are faulty.