Thanks to substantial past and recent developments, model predictive control has become one of the most relevant advanced control techniques. Nevertheless, many challenges associated to the reliance of MPC on a mathematical model that accurately depicts the controlled process still exist. This thesis is concerned with three of these challenges, placing the focus on constructing mathematically sound MPC controllers that are comparable in complexity to standard MPC implementations. The first part of this thesis tackles the challenge of model uncertainty in time-varying plants. A new dual MPC controller is devised to robustly control the system in presence of parametric uncertainty and simultaneously identify more accurate representations of the plant while in operation. The main feature of the proposed dual controller is the partition of the input, in order to decouple both objectives. Standard robust MPC concepts are combined with a persistence of excitation approach that guarantees the closed-loop data is informative enough to provide accurate estimates. Finally, the adequacy of the estimates for updating the MPC's prediction model is discussed. The second part of this thesis tackles a specific type of time-varying plant usually referred to as switching systems. A new approach to the computation of dwell-times that guarantee admissible and stable switching between mode-specific MPC controllers is proposed. The approach is computationally tractable, even for large scale systems, and relies on the well-known exponential stability result available for standard MPC controllers. The last part of this thesis tackles the challenge of MPC for large-scale networks composed by several subsystems that experience dynamical coupling. In particular, the approach devised in this thesis is non-cooperative, and does not rely on arbitrarily chosen parameters, or centralized initializations. The result is a distributed control algorithm that requires one step of communication between neighbouring subsystems at each sampling time, in order to properly account for the interaction, and provide admissible and stabilizing control.

The Intensive Care Unit (ICU) is where the critically-ill are treated. The first 24-hours ('the golden hours') of treatment is crucial to determine patient's recovery and survival, and mechanical ventilation plays a major role as the main life support system in the ICU. The efficiency of mechanical ventilation and its management strategy are assessed by observing the arterial blood gases (ABG), which are sampled every few hours using a catheter inserted into the patient's artery. This procedure is invasive thus can only be performed a handful of times each day. The ICU also has an abundance of underutilized data which until recently can only be translated by expert clinicians, who unfortunately always have clinical responsibilities to undertake concomitantly. This thesis proposes a series of new fuzzy logic-based models with a new type of fuzzy sets (type-2), which have not been investigated before in this clinical setting, for the relative dead-space (Kd), the carbon-dioxide production (VCO2), and the shunt sub-components for the SOPAVent (Sheffield Simulation of Patients under Artificial Ventilation) system, which performs predictions of arterial blood gases non-invasively and automatically. The Kd model, the VCO2 model and the resulting overall SOPAVent model are validated with retrospective real ICU patient data obtained from the Sheffield Royal Hallamshire Hospital (UK). The SOPAVent model is also validated with newly obtained data from patients diagnosed with Faecal Peritonitis (FP), from the Sheffield Royal Hallamshire Hospital (UK). Results showed an improved prediction accuracy for the Kd and the VCO2 sub-components when compared to existing systems. The prediction capability of SOPAVent is also improved from previous models for arterial blood gases before and after ventilator settings changes are made. A second new simplified model for predicting ABG using ventilator settings is also proposed with excellent prediction outcomes. Additionally, this thesis also looks into Electrical Impedance Tomography (EIT) as a potential bedside monitoring tool for pulmonary functions. EIT has the ability to provide a non-invasive, portable, and a relatively low cost alternative to other medical imaging systems. This thesis details the development of the hardware for a compact 16-electrode EIT measurement system, with the objective for future pulmonary applications. A method to generate three-dimensional (3D) images of the lungs from two-dimensional (2D) medical images of the thorax is also proposed with the estimation of lung volumes being presented.

No description available.

No description available.

No description available.

No description available.

No description available.

No description available.

No description available.

No description available.