Control-Aware Distributed Channel Access for Networked Control Systems

Modern industrial control environments consist of a multitude of spatially distributed sensors and actuators that form various complex control loops. Networked control systems (NCSs) refer to such systems wherein the communication and information exchange between these distributed components happen over a shared communication network. NCSs offer several advantages over their traditional counterparts, such as higher flexibility and scalability with lower deployment and maintenance costs. However, they also introduce novel challenges that need to be addressed before their full potential can be realized. A major challenge is the limited capacity of the network which necessitates efficient sharing of the available communication resources among control loops based on their real-time needs. In this thesis, we first focus on wired NCSs and advocate using control-aware priority measures by using the concepts of cost of information loss (CoIL) and value of information (VoI). Another challenge is orchestrating channel access without a central network manager which we address by proposing a distributed priority-based channel access method. Next, we turn our attention to wireless NCSs (WNCSs) and consider memoryless erasure channels. Despite the additional complexities that arise due to the unreliability of channels, we prove that the aforementioned priority measures can be utilized for control-aware distributed channel access. We further investigate the stability of the system under the proposed channel access scheme and establish stability conditions. Next, we consider scenarios that necessitate the use of learning methods to evaluate transmission priorities and address how the learning can be done in a distributed way without compromising control performance. Data transmission in certain industrial environments is prone to correlated packet dropouts, which cannot be modeled by the memoryless channel. We investigate such scenarios by considering a more comprehensive channel model, i.e., a two-state Markov chain, and derive the resulting priority measures for control-aware channel access in such settings. The priorities are proved to be a function of the transition probabilities of the underlying channel model which can be unknown in practice. We thus consider practical scenarios without a priori knowledge of these parameters and develop a model-based Bayesian reinforcement learning algorithm to learn them in a distributed control-aware manner