Search

Cyber Physical Systems (CPS) are complex hybrid systems integrating physics-based and digital worlds. Compared to the Internet of Things (IoT), CPS applications often have many complex physical behaviours, which makes the solutions monolithic and complex; thus making the developed CPS applications difficult to reuse. Furthermore, the classical isolated model-based design can be complicated when developing decentralised CPS applications. To lower the entry-barrier for developing the CPS applications, this thesis proposes an Open Resource-Oriented Architecture (OROA) to build the infrastructures to develop further varied CPS applications. Stakeholders, like developers, can exploit the advantages from the infrastructures for some features like low entry barrier, scalability and interoperability. This work is inspired by the REST (REpresentational State Transfer) architectural style, which is the network-based software architecture designed for the web. In recent years, the REST architectural style is also widely used in IoT applications since many of these have similar requirements as web applications. However, REST architectural style has many limitations when dealing with physical environments, because most current IoT applications are data-centric without complex physical behaviours. To meet these challenges, we discussed lacking design principles and features in the REST architectural style to develop the CPS applications such as abstracting ability, uncertainty handling and access control. We proposed a solution for each design principle, and eventually used the resource concept and Semantic Web technologies to accommodate all the requirements and implement the Rinfra framework to build the infrastructures for decentralised CPS applications. To cooperate with the designed RInfra (Resource Infrastructure), Constrained Application Protocol (CoAP) was extended to support context-adaptation. Scenarios in the smart transport and smart building have been presented as case studies to explain how the proposed architecture works and how it can benefit the decentralised CPS applications in heterogeneity, extensibility, interoperation and adaptation.

In the real world there are three types of multivariable control systems. The first one is when the number of inputs is equal to the number of the outputs, this type of multivariable control system is defined as a squared multivariable control system and the main type of controller designed is a decoupling controller which minimizes interactions and gives good set-point tracking. The second type of multivariable control system is where the number of inputs is greater than the number of the outputs, for this type of system the main controller designed is a fail-safe controller. This controller remains stable if a sub-set of actuator fail. The third type of multivariable control system is the number of outputs is greater than the number of inputs, for this type of system the main controller designed is an override control system. This controller only controls a sub-set of outputs based on a lowest wins control strategy. All the three types of multivariable control systems are included in this thesis. In this thesis the design of multivariable decoupling control, multivariable fail-safe control and multivariable override control as considered. The invention of evolutionary computing techniques has changed the design philosophy for control system design. Rather than using conventional techniques such as Nyquest plots or root-loci control systems can be designed using evolutionally algorithm. Such algorithms evolve solutions using cost functions and optimization. There are a variety of system performance indicators such as integral squared error operator has been used as cost functions to design controllers using such algorithms. The design of both fail-safe and override multivariable controllers is a difficult problem and there are very few analytical design methods for such controllers. Therefore, the main objective of this thesis is to use the genetic algorithms to involve both fail-safe and override controller multivariable controllers, such that they perform well in the time-domain.