Data quality assessment: Modelling and application in resilient monitoring systems

This paper presents a novel data quality model as part of a monitoring system that degrades gracefully under attacks on its sensors. The attacker is assumed to manipulate the sensor data's variance or mean, with the aim of projecting a false state of the plant. Each sensor's data is assigned a level of trust, termed data quality, as part of assessing the states of the process variables. For the variance-based attacker, it is established that the concept of data quality is not, in fact, necessary to obtain the best possible assessment. For the mean-based attacker, it is recognized that statistical means are not sufficient to discern data quality. To combat this problem, the so-called method of probing signals is proposed. The efficacy of this method is illustrated by numerical experiments categorized into two parts. The first deals with individual process variable assessment, while the second deals with the adaptation of the sensor network to obtain the best possible plant assessment.