Resilient Controller Design and Resource Allocation for Networked Control Systems

Networked control systems (NCS) include a communication network in the control loop. This results in network delay and packet dropping, and makes the system vulnerable to cyberattacks. The delays and packet dropping can become unacceptable if the quality of service (QoS) of the network deteriorates. This dissertation proposes a resilient controller design and a delay compensation algorithm to eliminate the adverse effects of time delay and cyber-attacks on the controller and observer in NCS. To maintain, acceptable QoS, three different bandwidth allocation strategies are proposed. The contributions of this dissertation are as follows, 1) Assuming that the cyberattack can be modeled as bounded perturbations on the controller and observer gains, a resilient observer and controller design using linear matrix inequality (LMI) is proposed to mitigate their effect. In addition, a time delay compensation algorithm is presented to eliminate the adverse effect of random time delays in NCS. The proposed resilient NCS design is applied to an active mass drive system subjected to several historical ground motions. 2) Three bandwidth allocation schemes are proposed for a NCS that shares the communication network with random traffic. In the first approach, the random traffic is predicted by employing Poisson graphical model. l_1 minimization is used to allocate bandwidth for an event-driven controller channel and a time-driven sensor channel. The sampling period for the time-driven channel is increased to reduce traffic to meet bandwidth constraints. In the second approach, a Hidden Markov Model (HMM) that includes a sufficient number of states is formulated using network simulation data and the Viterbi algorithm is used to predict the random traffic. Because the sampling period in the time-driven sensor channel is time-varying, a new stability condition for a linear time-invariant system with time-varying sampling period is presented. In the third approach, because the dynamic of the network has nonlinearities and uncertainties, network congestion is defined as a linear state space model which has perturbation. Q-learning based control is applied to the perturbed linear model by combining approximate dynamic programming and optimal control. To demonstrate the effectiveness of the proposed method, its results are compared to the results of the linear quadratic regulator