Distributed Fault Detection in Smart Spaces Based on Trust Management

Application performance in a smart space is affected by faulty behaviours of nodes and communication networks. Detection of faults helps diagnosis of problems and maintenance can be done to restore performance, for example, by replacing or reconfiguring faulty parts. Fault detection methods in the literature are too complex for typical low-resource devices and they do not perform well in detecting intermittent faults. We propose a fully distributed fault detection method that relies on evaluating statements about trustworthiness of aggregated data from neighbors. Given one or more trust statements that describe a fault-free state, the trustor node determines for each observation coming from the trustee whether it is an outlier or not. Several fault types can be explored using different trust statements whose parameters are assessed differently. The trustor subsequently captures the observation history of the trustee node in only two evidence variables using evidence update rules that give more weight to recent observations. The proposed method detects not only permanent faults but also intermittent faults with high accuracy and low false alarm rate.