Fault Detection and Isolation in Controlled Multi-Robot Systems

Taoufik, Anass (2021) Fault Detection and Isolation in Controlled Multi-Robot Systems. Doctoral thesis, Northumbria University.

Preview

Text (Doctoral Thesis) taoufik.anass_phd_18045909.pdf - Submitted Version Download (13MB) | Preview

Abstract

Multi-Agent Systems (MASs) have attracted much popularity, since the previous decade due to their potential wide range of applications. Indeed, connected MASs are deployed in order to achieve more complex objectives that could otherwise not be achievable by a single agent. In distributed schemes, agents must share their information with their neighbours, which are then used for common control and fault detection purposes, and thus do not require any central monitoring unit. This translates into the necessity to develop efficient distributed algorithms in terms of robustness and safety. Indeed, the problem of safety in connected cooperative MASs has arisen as a consequence of their complexity and the nature of their operations and wireless communication exchanges, which renders them vulnerable to not only physical faults, but also to cyber-attacks. The main focus of this thesis is the study of distributed fault and attack detection and isolation in connected MASs. First, a distributed methodology for global detection of actuator faults in a class of linear MASs with unknown disturbances is proposed using a cascade of fixed-time Sliding Mode Observers (SMOs), where each agent having access to their state, and neighbouring information exchanges, can give an exact estimate of the state of the overall MAS. An LMI-based approach is then applied to design distributed global robust residual signals at each agent capable of detecting faults anywhere in the network. This is then extended to agents with nonlinear nonholonomic dynamics where a new distributed robust Fault Detection and Isolation (FDI) scheme is proposed using predefined-time stability techniques to derive adequate distributed SMOs. This enables to reconstruct the global system state in a predefined-time and generate proper residual signals. The case of MASs with higher order integrator dynamics, where only the first state variable is measurable and the topology is switching is investigated, where a new approach to identify faults and deception attacks is introduced. The proposed protocol makes an agent act as a central node monitoring the whole system activities in a distributed fashion whereby a bank of distributed predefined-time SMOs for global state estimation are designed, which are then used to generate residual signals capable of identifying cyber-attacks despite the switching topology. The problem of attack and FDI in connected heterogeneous MASs with directed graphs, is then studied. First, the problem of distributed fault detection for a team of heterogeneous MASs with linear dynamics is investigated, where a new output observer scheme is proposed which is effective for both directed and undirected topologies. The main advantage of this approach is that the design, being dependant only on the input-output relations, renders the computational cost, information exchange and scalability very effective compared to other FDI approaches that employ the whole state estimation of the agents and their neighbours as a basis for their design. A more general model is then studied, where actuator, sensor and communication faults/attacks are considered in the robust detection and isolation process for nonlinear heterogeneous MASs with measurement noise, dynamic disturbances and communication parameter uncertainties, where the topology is not required to be undirected. This is done using a distributed finite-frequency mixed H_/H1 nonlinear UIO-based approach. Simulation examples are given for each of the proposed algorithms to show their effectiveness and robustness.

Actions (login required)

View Item

Downloads