No abstract

Resilient monitoring systems, considered in this paper, are sensor networks that degrade gracefully under malicious attacks on their sensors, causing them to project misleading information. The goal of this paper is to design, analyze, and evaluate the performance of a resilient monitoring system intended to monitor plant conditions (normal or anomalous). The architecture developed consists of four layers: data quality assessment, process variable assessment, plant condition assessment, and sensor network adaptation. Each of these layers is analyzed by either analytical or numerical tools. The performance of the overall system is evaluated using a simplified boiler/turbine plant. The measure of resiliency is quantified based on the Kullback-Leibler divergence and shown to be sufficiently high in all scenarios considered.

No abstract

Resilient monitoring systems (RMSs) are sensor networks that degrade gracefully under cyber-attacks on their sensors. The recently developed RMSs, while being effective in the attacked sensors identification and isolation, exhibited a drawback in their operation-an exponentially increasing assessment time as a function of the number of sensors in the network. To combat this curse of dimensionality, a decomposition approach has been proposed, which led to a dramatic reduction of the assessment time, irrespective of the sensor network dimensionality. However, information losses and, thus, reductions in the level of resiliency due to the decomposition, have not been investigated. This paper is intended to carry out such an investigation. Specifically, it derives conditions for lossless decomposition in terms of the Renyi-2 entropy. The development is based on the analysis of matrices, which characterize coupling of process variables and on a monotonicity property of the Dempster-Shafer combination rule on a class of functions, which arise within the RMS operation.