<p>Model Predictive Control (MPC) is traditionally designed assuming no model mismatch and tuned to provide acceptable behavior when mismatch occurs. This thesis extends the MPC design to account for explicit mismatch in the control and optimization of a wide range of uncertain dynamic systems with feedback, such as in process control and supply chain optimization.</p> <p>The major contribution of the thesis is the development of a new MPC method for robust performance, which offers a general framework to optimize the uncertain system behavior in the closed-loop subject to hard bounds on manipulated variables and soft bounds on controlled variables. This framework includes the explicit handling of correlated, time-varying or time-invariant, parametric uncertainty appearing externally (in demands and disturbances) and internally (in plant/model mismatch) to the control system. In addition, the uncertainty in state estimation is accounted for in the controller.</p> <p> For efficient and reliable real-time solution, the bilevel stochastic optimization formulation of the robust MPC method is approximated by a limited number of (convex) Second Order Cone Programming (SOCP) problems with an industry-proven heuristic and the classical chance-constrained programming technique. A closed-loop uncertainty characterization method is also developed which improves real-time tractability by performing intensive calculations off-line.</p> <p>The new robust MPC method is extended for process control problems by integrating a robust steady-state optimization method addressing closed-loop uncertainty. In addition, the objective function for trajectory optimization can be formulated as nominal or expected dynamic performance. Finally, the method is formulated in deviation variables to correctly estimate time-invariant uncertainty.</p> <p>The new robust MPC method is also tailored for supply chain optimization, which is demonstrated through a typical industrial supply chain optimization problem. The robust MPC optimizes scenario-specific safety stock levels while satisfying customer demands for time-varying systems with uncertainty in demand, manufacturing and transportation. Complexity analysis and computational study results demonstrate that the robust MPC solution times increase with system scale moderately, and the method does not suffer from the curse of dimensionality.</p> / Thesis / Doctor of Philosophy (PhD)

We propose a novel framework for motion planning and control that is based on a manifold encoding of the desired solution set. We present an alternate, model-free, approach to path planning, replanning and control. Our approach is founded on the idea of encoding the set of possible trajectories as a skill manifold, which can be learnt from data such as from demonstration. We describe the manifold representation of skills, a technique for learning from data and a method for generating trajectories as geodesics on such manifolds. We extend the trajectory generation method to handle dynamic obstacles and constraints. We show how a state metric naturally arises from the manifold encoding and how this can be used for reactive control in an on-line manner. Our framework tightly integrates learning, planning and control in a computationally efficient representation, suitable for realistic humanoid robotic tasks that are defined by skill specifications involving high-dimensional nonlinear dynamics, kinodynamic constraints and non-trivial cost functions, in an optimal control setting. Although, in principle, such problems can be handled by well understood analytical methods, it is often difficult and expensive to formulate models that enable the analytical approach. We test our framework with various types of robotic systems – ranging from a 3-link arm to a small humanoid robot – and show that the manifold encoding gives significant improvements in performance without loss of accuracy. Furthermore, we evaluate the framework against a state-of-the-art imitation learning method. We show that our approach, by learning manifolds of robotic skills, allows for efficient planning and replanning in changing environments, and for robust and online reactive control.

Unmanned Aerial Vehicles (UAVs) are remotely piloted or self-piloted aircrafts that can carry cameras, sensors, communications equipment or other payloads. Tiltrotor UAVs provide a unique platform that fulfills the needs for ever-changing mission requirements by combining the desired features / hovering like a helicopter and reaching high forward speeds like an airplane. In this work, the conceptual design and aerodynamical model of a realizable small-sized Tiltrotor UAV is constructed, the linearized state-space models are obtained around the trim points for airplane, helicopter and conversion modes, controllers are designed using Linear Quadratic Regulator (LQR) methods and gain-scheduling is employed to obtain a simulation for the whole flight envelope. The ideas for making a real flying model are established according to simulation results.

No description available.

The energy retrieval process in an Enhanced Geothermal System (EGS) depends on fracture networks to facilitate fluid movement, thereby enabling the extraction of heat from adjacent rocks matrix. Nonetheless, due to the inherent heterogeneity and intricate multi-physics characteristics of these systems, high-fidelity physics-based forward simulations ($f_h$) can be computationally demanding. This presents a considerable obstacle to the efficient management of these reservoirs. Therefore, creating an effective and robust optimization framework is essential, with the primary aim being to maximize the thermal extraction from Enhanced Geothermal Systems (EGS). A deep learning-assisted reservoir management framework incorporating a low-fidelity forward surrogate model ($f_l$) alongside gradient-based optimizers is developed to expedite reservoir management. A thermo-hydro-mechanical (THM) model for EGS is established by utilizing finite element-based reservoir simulation techniques. By parameterizing the fracture aperture and well controls, we carried out the THM simulation to produce 2500 datasets. Subsequently, we employed these datasets to train two distinct deep neural network (DNN) architectures to predict the variations in pressure and temperature distributions. Ultimately, these predictions from the forward model are used in calculating the total net energy. Instead of executing the optimization workflow with a large number of simulations from $f_h$, we directly optimize the well control parameters relative to the geological parameters using $f_l$. Since $f_l$ is efficient and fully differentiable, it could be combined with various gradient-based or gradient-free optimization algorithms to maximize the total net energy by determining the optimal decision parameters. Drawing from the simulation datasets, we analysed the effect of fracture aperture variation on temperature and pressure evolution. Our investigation revealed that the spatial distribution of the fracture aperture is a predominant factor in controlling the propagation of the thermal front. Variations of the fracture aperture exhibit a strong correlation with temperature fluctuations within the fracture, primarily due to thermal stress changes. When compared with a comprehensive physics simulator, our DNN-based forward surrogate model offers a significant computational acceleration, approximately 1500 times faster, without compromising predictive accuracy, achieving an $R^2$ value of 99%. The forward model $f_l$, when combined with gradient-based optimizers, enables optimization to proceed 10 to 68 times faster than when using derivative-free global optimizers. The proposed reservoir management framework exhibits both efficiency and scalability, facilitating the real-time execution of each optimization process.

No description available.

During the last 10 years solar cooling systems attracted more and more interest not only in the research area but also on a private and commercial level. Several demonstration plants have been installed in different European countries and first companies started to commercialise also small scale absorption cooling machines. However, not all of the installed systems operate efficiently and some are, from the primary energy point of view, even worse than conventional systems with a compression chiller. The main reason for this is a poor system design combined with suboptimal control. Often several non optimised components, each separately controlled, are put together to form a ‘cooling system’. To overcome these drawbacks several attempts are made within IEA task 38 (International Energy Agency Solar Heating and Cooling Programme) to improve the system design through optimised design guidelines which are supported by simulation based design tools. Furthermore, guidelines for an optimised control of different systems are developed. In parallel several companies like the SolarNext AG in Rimsting, Germany started the development of solar cooling kits with optimised components and optimised system controllers. To support this process the following contributions are made within the present work: - For the design and dimensioning of solar driven absorption cooling systems a detailed and structured simulation based analysis highlights the main influencing factors on the required solar system size to reach a defined solar fraction on the overall heating energy demand of the chiller. These results offer useful guidelines for an energy and cost efficient system design. - Detailed system simulations of an installed solar cooling system focus on the influence of the system configuration, control strategy and system component control on the overall primary energy efficiency. From the results found a detailed set of clear recommendations for highly energy efficient system configurations and control of solar driven absorption cooling systems is provided. - For optimised control of open desiccant evaporative cooling systems (DEC) an innovative model based system controller is developed and presented. This controller consists of an electricity optimised sequence controller which is assisted by a primary energy optimisation tool. The optimisation tool is based on simplified simulation models and is intended to be operated as an online tool which evaluates continuously the optimum operation mode of the DEC system to ensure high primary energy efficiency of the system. Tests of the controller in the simulation environment showed that compared to a system with energy optimised standard control the innovative model based system controller can further improve the primary energy efficiency by 19 %.

Differential drag controllers have been examined in the context of a full system simulation of a target/chaser pair of spacecraft in low Earth orbit. An Extended Kalman Filter has been designed to process measurement sets from GPS receivers on the target and chaser spacecraft. The estimated state from the Kalman Filter is used in a differential drag control algorithm to determine the appropriate control action. Modifications are made to the standard differential drag control algorithms to reduce unnecessary actuations in the presence of errors in the dynamic modeling, control actuation, and incoming measurements. Detailed explanations of the algorithms, dynamic models, and derivations for both the Kalman Filter and the differential drag control laws are presented. Multiple test cases are used to validate the controller performance under a variety of initial conditions. In these simulations, the differential drag control algorithms successfully maneuver the chaser spacecraft from the initial conditions to a final state with instantaneous time-average position (relative to the target spacecraft) of not more than 10 meters in the radial and in-track directions. Modifications to the standard control algorithms ensure that extraneous control actuations are minimized. An optimization algorithm is used determine the time-optimal differential drag control history, and the results are compared to the standard control logic and modified control logic. Based on the optimization results, it is recommended that a system employing differential drag control (especially those with limited computational resources) should use the modified control logic, as it provides a standardized methodology that can be followed in any mission. / text

In this thesis, we develop scalable frameworks to (re)design a class of large-scale network systems with built-in control mechanisms, including electric power systems and the Internet, in order to improve their economic efficiency and performance while guaranteeing their stability and robustness. After a detailed introduction relating to power system control and optimization, as well as network congestion control, we turn our attention to merging primary and secondary frequency control for the power grid. We present modifications in the conventional generation control using a consensus design approach while considering the participation of controllable loads. The optimality, stability and delay robustness of the redesigned system are studied. Moreover, we extend the proposed control scheme to (i) networks with more complexity and (ii) the case where controllable loads are involved in the optimization. As a result, our controllers can balance power flow and drive the system to an economically optimal operating point in the steady state. We then study a real-time control framework that merges primary, secondary and tertiary frequency control in power systems. In particular, we consider a transmission level network with tree topology. A distributed dynamic feedback controller is designed via a primal-dual decomposition approach and the stability of the overall system is studied. In addition, we introduce extra dynamics to improve system performance and emphasize the trade-off when choosing the gains of the extra dynamics. As a result, the proposed controller can balance supply and demand in the presence of disturbances, and achieve optimal power flow in the steady state. Furthermore, after introducing the extra dynamics, the transient performance of the system significantly improves. A redesign framework for network congestion control is developed next. Motivated by the augmented Lagrangian method, we introduce extra terms to the Lagrangian, which is used to redesign the primal-dual, primal and dual algorithms. We investigate how the gains resulting from the extra dynamics influence the stability and robustness of the system. Moreover, we show that the overall system can achieve added robustness to communication delays by appropriately tuning these gains. Also, the meaning of these extra dynamics is investigated and a distributed proportional-integral-derivative controller for solving network congestion control problems is further developed. Finally, we concentrate on a reverse- and forward-engineering framework for distributed control of a class of linear network systems to achieve optimal steady-state performance. As a typical illustration, we use the proposed framework to solve the real-time economic dispatch problem in the power grid. On the other hand, we provide a general procedure to modify control schemes for a special class of dynamic systems. In order to investigate how general the reverse- and forward-engineering framework is, we develop necessary and sufficient conditions under which an linear time-invariant system can be reverse-engineered as a gradient algorithm to solve an optimization problem. These conditions are characterized using properties of system matrices and relevant linear matrix inequalities. We conclude this thesis with an account for future research.

No description available.