Missing data are common wherever statistical methods are applied in practice. They present a problem by demanding that additional untestable assumptions be made about the mechanism leading to the incompleteness of the data. Minimising the strength of these assumptions and assessing the sensitivity of conclusions to their possible violation constitute two important aspects of current research in this area. One attractive approach is the doubly robust (DR) weighting-based method proposed by Robins and colleagues. By incorporating two models for the missing data process, inferences are valid when at least one model is correctly specified. The balance between robustness, efficiency and analytical complexity is one which is difficult to strike, resulting in a split between the likelihood and multiple imputation (MI) school on one hand and the weighting and DR school on the other. We propose a new method, doubly robust multiple imputation (DRMI), combining the convenience of MI with the robustness of the DR approach, and explore the use of our new estimator for non-monotone missing at random data, a setting in which, hitherto, estimators with the DR property have not been implemented. We apply the method to data from a clinical trial comparing type II diabetes drugs, where we also use MI as a tool to explore sensitivity to the missing at random assumption. Finally, we study DRMI in the longitudinal binary data setting and find that it compares favourably with existing methods.

This thesis describes a new agent-based architecture called the Limited Instruction Set Agent (LISA). Agent-based systems are a popular approach to the implementation of autonomous behaviour, and they usually consist of a 'reasoning' module that commands lower level subsystems that in turn interact with the environment. When an autonomous system is placed in any environment, the correctness of the software must be guaranteed for safety. This is generally done with 'verification by model checking' which consists of creating a model, which represents the system and its interaction with the environment, and then proving specifications using the model. Most agent frameworks to date do not contemplate verification as a design feature and they generally share a few drawbacks: the generation of a model that can be verified by a model checking software is either done manually or by executing the agent code recursively and exploring every possible path to list the state space of the system. The LISA system is based on existing agent-based architectures and it is designed to be structurally simpler than its predecessors with the aim of facilitating the verification process. The agent program of LISA is enriched with structures that allow to model the probabilistic nature of environmental events, so that they can be taken into account in the verification process. The LISA program can be automatically translated to a verifiable probabilistic model suitable for verification with existing software tools such as PRISM. Furthermore, the system is structured to minimise the size of its probabilistic model, and ultimately offers a faster verification process. The thesis contains a number of theoretical contributions to the LISA programming system, including run-time verification for prediction of future outcomes of actions, and the new methods are illustrated on the programming and simulation with an example of autonomous surface vehicle for sea mine detection and disposal.

In the last five decades, exoskeletons have emerged as a solution to assist paraplegic and elderly patients perform fundamental mobility tasks. The main challenge nowadays, is to develop a device that is safe, power sufficient, seamlessly integrates with the user, while being affordable. Several solutions have been proposed, and controllers have been identified as the only component which can enhance integration with the user without adding weight to the system, or increasing energy consumption. Moreover, a software platform where the mechanical design and control techniques can be assessed, prior to experimental trials, could save resources and decrease costs. In this thesis, the development of humanoid and exoskeleton models, within the SimWise virtual environment, to perform an initial validation of controllers proposed without the need of a physical prototype, is performed. Furthermore, the selected platform is evaluated regarding its fitness for this application. The methodology used to generate CAD models of a humanoid, exoskeletons and a wheel walker within the SimWise virtual environment is described, along with its integration with MATLAB Simulink. Two exoskeleton models with their corresponding controllers were developed, firstly, a hybrid exoskeleton with a wheel walker for restoration of walking in paraplegic patients. And secondly, an actuated exoskeleton for assistance in standing-up and sitting-down motions in both the elderly and paraplegic patients. The hybrid exoskeleton uses functional electrical stimulation as actuation for knee joints and a frame with brakes mounted at hip, knee and ankle joints to generate the walking cycle. The wheel walker is used for support and equilibrium. A fuzzy controller for the low level and a finite state controller for the middle level is developed. Validation of the system over repeated walking cycles, including external disturbances, and simulation of use by humanoids of different dimensions, is performed within the virtual environment and results discussed. PID low level control of hip and knee joints is used to analyse standing-up and sittingdown motions, and incorporated with an actuated exoskeleton for assisting elderly people on performing the aforementioned tasks. A finite state middle level control is developed to generate reference trajectories at variable velocities for the restoration of these motions for paraplegic patients. An optimisation algorithm is used to identify low level controller parameters for ankle joints. Finally, offline and online calculation and incorporation of zero moment point in the control loop is performed to assess equilibrium of the system.

System Identification (SID) is an important area of structural dynamics and is concerned with constructing a functional relationship between the inputs and the outputs of a system. Furthermore, it estimates the parameters that the studied system depends upon. This aspect of structural dynamics has been studied for many years and computational methods have been developed in order to deal with the system identification of real structures, with the aim of getting a better understanding of their dynamic behaviour. The most straightforward classification of structures is into structures that behave linearly and structures that behave nonlinearly. Even so, one needs to keep in mind that no structure is indefinitely linear. During its service, a structure can behave nonlinearly at any given point, under the right working and environmental conditions. A key challenge in applying SID to real systems is in handling the uncertainty inherent in the process. Uncertainty arises from various sources such as modelling error and measurement error (noisy data), resulting in uncertainty in the parameter estimates. One of the ways in which one can deal with uncertainty is by adopting a probabilistic framework. In this way one admits the limitations in the process of SID through providing probability distributions over the models and parameters of interest, rather than a simple 'best estimate'. Throughout this work a Bayesian probabilistic framework is adopted as it covers the uncertainty issue without over-fitting (it provides the simplest, least complex solutions to the issue at hand). Of great interest when working within a Bayesian framework are Markov Chain Monte Carlo(MCMC) sampling methods. Of relevance to this research are the Metropolis-Hastings(MH) algorithm and the Reversible Jump Markov Chain Monte Carlo(RJMCMC) algorithm. Markov Chain Monte Carlo(MCMC) methods and algorithms have been extensively investigated for linear dynamical systems. One of the advantages of these methods being used in a Bayesian framework is that they handle uncertainty in a principled way. In recent years, increasing attention has been paid to the role nonlinearity plays in engineering problems. As a result, there is an increasing focus on developing computational tools that may be applied to nonlinear systems as well as linear systems, with the objective that they should provide reliable results at reasonable computational costs. The application of MCMC methods in nonlinear system identi cation(NLSID) has focused on parameter estimation. However, often enough, the model form of systems is assumed known which is not the case in many contexts(such as NLSID when the nonlinearity is hard to identify and model, or Structural Health Monitoring when the damage extent or number of damage sites is unknown). The current thesis is concerned with the development of computational tools for performing System Identification in the context of structural dynamics, for both linear and nonlinear systems. The research presented within this work will demonstrate how the Reversible Jump Markov Chain Monte Carlo algorithm, within a Bayesian framework, can be used in the area of SID in a structural dynamics context for doing both parameter estimation and model selection. The performance of the RJMCMC algorithm will be benchmarked throughout against the MH algorithm. Several numerical case studies will be introduced to demonstrate how the RJMCMC algorithm may be applied in linear and nonlinear SID; and one numerical case study to demonstrate application to a SHM problem. These will be followed by experimental case studies to evaluate linear and nonlinear SID performance for a real structure.

This thesis considers a number of important practical issues in the synthesis of discrete-time robust controllers for industrial processes. The work focuses on the control of an "unknown" SISO process (the IFAC 1993 benchmark), the design of robust model-based controllers for a MIMO industrial production process (a glass tube production process), and the design of robust MIMO controllers having integrated control and diagnostic capabilities. The industrial case studies presented are realistic in the sense that their control problems do frequently arise in engineering situations. Explicit state-space formulae for Hinfinity-based one degree-of-freedom (1-DOF) and two degrees-of-freedom (2-DOF) robust controllers are derived. They provide robust stability with respect to left coprime factor perturbations, and for the 2-DOF case, a degree of robust performance in the sense of making the closed-loop system follow a desired reference model. Robust controllers for the "unknown" plant are designed using H2 and Hinfinity optimization techniques. Explicit closed-loop performance is obtained by designing the weighting function parameters using numerical optimization techniques in the form of the method of inequalities. Methods for designing Hinfinity-based controllers that can be directly implemented in the Internal Model Control (IMG) scheme are presented. Explicit state-space formulae for Hinfinity-based IMG 1-DOF and 2-DOF robust controllers which provide robust stability and robust performance with respect to left coprime factor perturbations, axe derived. A technique for discrete-time model reduction is presented, with two illustrative examples. The technique is used in a detailed study of the identification and control of the glass tube production process. The production process, especially for large tube measures, is ill-conditioned and contains large time delays. The model of the process reflects the transfer of two process inputs (mandrel pressure and drawing speed) to the tube dimensions (wall thickness and diameter). The models obtained from advanced multivariable identification are used for the design of robust IMG controllers for the process. The robust performance of the controller is demonstrated and a comparison is made with the present control system. Finally, a framework for synthesizing robust controllers which have both control and actuator failure detection capabilities is presented. Simulation results for a MIMO design example are presented which demonstrate the feasibility of this integrated design approach.

No description available.

This thesis is interested in improving the operation of a parabolic trough technology based solar thermal power plant by means of automatic control. One of the challenging issues in a solar thermal power plant, from the control point of view, is to maintain the thermal process variables close to their desired levels. In contrast to a conventional power plant where fuel is used as the manipulated variable, in a solar thermal power plant, solar radiation cannot be manipulated and in fact it ironically acts as a disturbance due to its change on a daily and seasonal basis. The research facility ACUREX is used as a test bed in this thesis. ACUREX is a typical parabolic trough technology based solar thermal power plant and belongs to the largest research centre in Europe for concentrating solar technologies, namely the Plataforma Solar de Almería (PSA) in south-east Spain. The plant exhibits nonlinearities as well as resonance characteristics that lie well within the desired control bandwidth. Failure to adequately capture the resonance characteristics of the plant results in an undesired oscillatory control performance. Moreover, measured disturbances are an integral part of the plant and while some of the measured disturbances do not have a significant impact on the operation of the plant, others do. Hence, with the aim of handling the plant nonlinearities and capturing the plant resonance characteristics, while taking explicit account of the measured disturbances, in this thesis a gain scheduling feedforward predictive control strategy is proposed. The control strategy is based upon a family of local linear time-invariant state space models that are estimated around a number of operating points. The locally estimated linear time-invariant state space models have the key novelty of being able to capture the resonance characteristics of the plant with the minimal number of states and hence, simple analysis and control design. Moreover, while simple classical, series and parallel, feedforward configurations have been proposed and used extensively in the literature to mitigate the impact of the measured disturbances of the ACUREX plant, the proposed control strategy incorporates a feedforward systematically by including the effects of the measured disturbances of the ACUREX plant into the predictions of future outputs. In addition, a target (set point) for a control strategy is normally set at the ACUREX plant by the plant operator. However, in this thesis it is argued that, in parallel, the operator must choose between potentially ambitious and perhaps unreachable targets and safer targets. Ambitious targets can lead to actuator saturation and safer targets imply electricity production losses. Hence, in this thesis a novel two-layer hierarchical control structure is proposed with the gain scheduling feedforward predictive control strategy being deployed in a lower layer and an adequate reachable reference temperature being generated from an upper layer. The generated reference temperature drives the plant near optimal operating conditions, while satisfying the plant safety constraints, without any help from the plant operator and without adding cost. The proposed two-layer hierarchical control strategy has the potential benefits of: (i) maximising electricity production; (ii) reducing the risk of actuator saturation; (iii) extending the life span of various elements of the plant (e.g. synthetic oil, pump and valves) and (iv) limiting the role of the plant operator. The efficacy of the proposed two-layer hierarchical control strategy is evaluated using a nonlinear simulation model that approximates the dynamic behaviour of the ACUREX plant. The nonlinear simulation model is constructed in this thesis and validated in the time and frequency domain.

No description available.

Swarm robotic systems comprise many individual robots, and exhibit a degree of innate fault tolerance due to this built-in redundancy. They are robust in the sense that the complete failure of individual robots will have little detrimental effect on a swarm's overall collective behaviour. However, it has recently been shown that partially failed individuals may be harmful, and cause problems that cannot be solved by simply adding more robots to the swarm. Instead, an active approach to dealing with failed individuals is required for a swarm to continue operation in the face of partial failures. This thesis presents a novel method of exogenous fault detection that allows robots to detect the presence of faults in each other, via the comparison of expected and observed behaviour. Each robot predicts the expected behaviour of its neighbours by simulating them online in an internal replica of the real world. This expected behaviour is then compared against observations of their true behaviour, and any significant discrepancy is detected as a fault. This work represents the first step towards a distributed fault detection, diagnosis, and recovery process that would afford robot swarms a high degree of fault tolerance, and facilitate long-term autonomy.

This thesis presents research carried out as part of a project undertaken in fulfilment of the requirements of Loughborough University for the award of Philosophical Doctorate. The main focus of this research is to investigate and develop an appropriate level of automation to the existing manual finishing operations of small metallic components to achieve required surface quality and to remove superficial defects. In the manufacturing industries, polishing processes play a vital role in the development of high precision products, to give a desired surface finish, remove defects, break sharp edges, extend the working life cycle, and meet mechanical specification. The polishing operation is generally done at the final stage of the manufacturing process and can represent up to a third of the production time. Despite the growth automated technology in industry, polishing processes are still mainly carried out manually, due to the complexity and constraints of the process. Manual polishing involves a highly qualified worker polishing the workpiece by hand. These processes are very labour intensive, highly skill dependent, costly, error-prone, environmentally hazardous due to abrasive dust, and - in some cases - inefficient with long process times. In addition, the quality of the finishing is dependent on the training, experience, fatigue, physical ability, and expertise of the operator. Therefore, industries are seeking alternative solutions to be implemented within their current processes. These solutions are mainly aimed at replacing the human operator to improve the health and safety of their workforce and improve their competitiveness. Some automated solutions have already been proposed to assist or replace manual polishing processes. These solutions provide limited capabilities for specific processes or components, and a lack of flexibility and dexterity. One of the reasons for their lack of success is identified as neglecting the study and implementing the manual operations. This research initially hypothesised that for an effective development, an automated polishing system should be designed based on the manual polishing operations. Therefore, a successful implementation of an automated polishing system requires a thorough understanding of the polishing process and their operational parameters. This study began by collaborating with an industrial polishing company. The research was focused on polishing complex small components, similar to the parts typically used in the aerospace industry. The high level business processes of the polishing company were capture through several visits to the site. The low level operational parameters and the understanding of the manual operations were also captured through development of a devices that was used by the expert operators. A number of sensors were embedded to the device to facilitate recording the manual operations. For instance, the device captured the force applied by the operator (avg. 10 N) and the cycle time (e.g. 1 pass every 5 sec.). The capture data was then interpreted to manual techniques and polishing approaches that were used in developing a proof-of-concept Integrated Robotic Polishing System (IRPS). The IRPS was tested successfully through several laboratory based experiments by expert operators. The experiment results proved the capability of the proposed system in polishing a variety of part profiles, without pre-existing geometrical information about the parts. One of the main contributions made by this research is to propose a novel approach for automated polishing operations. The development of an integrated robotic polishing system, based on the research findings, uses a set of smart sensors and a force-position-by-increment control algorithm, and transpose the way that skilled workers carry out polishing processes.