Dynamics analysis and integrated design of real-time control systems

Real-time control systems are widely deployed in many applications. Theory and practice for the design and deployment of real-time control systems have evolved significantly. From the design perspective, control strategy development has been the focus of the research in the control community. In order to develop good control strategies, process modelling and analysis have been investigated for decades, and stability analysis and model-based control have been heavily studied in the literature. From the implementation perspective, real-time control systems require timeliness and predictable timing behaviour in addition to logical correctness, and a real-time control system may behave very differently with different software implementations of the control strategies on a digital controller, which typically has limited computing resources. Most current research activities on software implementations concentrate on various scheduling methodologies to ensure the schedulability of multiple control tasks in constrained environments. Recently, more and more real-time control systems are implemented over data networks, leading to increasing interest worldwide in the design and implementation of networked control systems (NCS). Major research activities in NCS include control-oriented and scheduling-oriented investigations. In spite of significant progress in the research and development of real-time control systems, major difficulties exist in the state of the art. A key issue is the lack of integrated design for control development and its software implementation. For control design, the model-based control technique, the current focus of control research, does not work when a good process model is not available or is too complicated for control design. For control implementation on digital controllers running multiple tasks, the system schedulability is essential but is not enough; the ultimate objective of satisfactory quality-of-control (QoC) performance has not been addressed directly. For networked control, the majority of the control-oriented investigations are based on two unrealistic assumptions about the network induced delay. The scheduling-oriented research focuses on schedulability and does not directly link to the overall QoC of the system. General solutions with direct QoC consideration from the network perspective to the challenging problems of network delay and packet dropout in NCS have not been found in the literature. This thesis addresses the design and implementation of real-time control systems with regard to dynamics analysis and integrated design. Three related areas have been investigated, namely control development for controllers, control implementation and scheduling on controllers, and real-time control in networked environments. Seven research problems are identified from these areas for investigation in this thesis, and accordingly seven major contributions have been claimed. Timing behaviour, quality of control, and integrated design for real-time control systems are highlighted throughout this thesis. In control design, a model-free control technique, pattern predictive control, is developed for complex reactive distillation processes. Alleviating the requirement of accurate process models, the developed control technique integrates pattern recognition, fuzzy logic, non-linear transformation, and predictive control into a unified framework to solve complex problems. Characterising the QoC indirectly with control latency and jitter, scheduling strategies for multiple control tasks are proposed to minimise the latency and/or jitter. Also, a hierarchical, QoC driven, and event-triggering feedback scheduling architecture is developed with plug-ins of either the earliest-deadline-first or fixed priority scheduling. Linking to the QoC directly, the architecture minimises the use of computing resources without sacrifice of the system QoC. It considers the control requirements, but does not rely on the control design. For real-time NCS, the dynamics of the network delay are analysed first, and the nonuniform distribution and multi-fractal nature of the delay are revealed. These results do not support two fundamental assumptions used in existing NCS literature. Then, considering the control requirements, solutions are provided to the challenging NCS problems from the network perspective. To compensate for the network delay, a real-time queuing protocol is developed to smooth out the time-varying delay and thus to achieve more predictable behaviour of packet transmissions. For control packet dropout, simple yet effective compensators are proposed. Finally, combining the queuing protocol, the packet loss compensation, the configuration of the worst-case communication delay, and the control design, an integrated design framework is developed for real-time NCS. With this framework, the network delay is limited to within a single control period, leading to simplified system analysis and improved QoC