Ph. D. Project
Fault Tolerant methods Design for a fleet of autonomous vehicules against faults/failures based on multi-agent systems
2019/05/27 - 2021/02/03
Context and Objectives: Distributed coordination of multiple networked dynamic agents has spurred a broad interest in the last decade. Compared to a solo complicated agent, greater efficiency, operational capability and low coats can be brought by networked multiple simpler and cheaper agents. To execute a common mission using networked multi-agent systems, the consistent agreement is required. This agreement is called "consensus" in multi-agent systems. The agreement variables are rendered with specific physical quantities in different environments, such as the workload in a network of parallel computers and the clock speed for wireless sensor networks. The "consensus algorithm" is a common iteration rule, which specifies the information exchange relationship and the instantaneous state update law for each single agent. To enable the capability of accomplishing a common work cooperatively, information consensus is crucial. The destination or heading direction in a cooperative mission with rendezvous task for instance is the information to be consensus. The information consensus is investigated in diverse fields due to its broad potential applications.
The main objective of our research is to systematically develop fault detection and isolation methods and the corresponding fault tolerant (FT) method. Unlike the classical FDI/FT problem, there is no centralized controller handling global information. The nodes acquire only local knowledge in networked multi-agent system (NMAS). Although recently some heuristic attempts are paving the approach to FDI/FT in NMAS, the fault diagnosis techniques in NMAS are still challenging. The following research objectives will be achieved through this research:
(1). Development of an effective residual generator for distributed multi-agent system
(2). Development of an unknown input decoupling strategy to better distinguishes the candidate faults
(3). Development of a systematic methodology for FDI/FT problem in distributed multi-agent system
(4). Application of the developed FDI/FT methods to multi-agent systems

Numerical simulations will be used to demonstrate the technological feasibility and effectiveness of the proposed methodologies. This simulation test bed relies on multi-modeling and co-simulation to integrated results from different scientific fields related to UAV.

Methodology: By ignoring the physical means associated with the specific agents, the individual agents can be abstracted as nodes in a graph. The interaction between agents can be encoded as edges connecting the corresponding nodes. Therefore, graph theory is formalized as a powerful tool for the research of networked multi-agent system. The underlying communication relationship among the agents is described by the topology of associated graph. Each agent dynamics is modeled as a linear system. The multiple agents share information under constraints. The constraint relationship is described by the associated communication topology. Since the agents may suffer from hardware malfunctions or software bugs while performing consensus seeking, the corresponding misbehaviours will thereby escalate due to the decentralized way of state updating. Model-based fault diagnosis techniques will be utilized to deal with the FDI problems in multi-agent system. The faulty agent is defined as the one without nominal consensus evolution. We will model misbehaviours as unknown inputs in each agent dynamics, by which the FDI problem will be cast into residual generation framework. An ideal residual generator can indicate the faulty agent if it is moving along a very unlikely trajectory. However, it is susceptible to some outside inputs, i.e. system uncertainties and external disturbances. To better distinguish the faulty signal from system uncertainties and external disturbances, the robust control theory will be recruited to bridge the gap between FDI theory and application. Theoretically, a residual generator can be perfectly decoupled from system uncertainties and external disturbances with enough number of sensors. However, this is not an economic way. Sometimes, adding redundant sensors is even impossible due to the space limitation. With the rapid growth of robust control theory, classical residual generator is better equipped in FDI techniques. Since the actualization of the robust residual generator is executed in distributed multi-agent system, the feasibility will be discussed before implementation. In this part, the skills and tools in robust control theory will be utilized, i.e. signal norms, singular value decomposition, coprime factorization, etc.
FTC, Fleet, collaborative control, Multi-agents system
Control Identification Diagnosis