Browsing by Subject "Process control"

Now showing 1 - 5 of 5
  • Thumbnail Image
    Item
    Design and control of integrated systems for hydrogen production and power generation
    (2013-11) Georgis, Dimitrios
    Growing concerns on CO2 emissions have led to the development of highly efficient power plants. Options for increased energy efficiencies include alternative energy conversion pathways, energy integration and process intensification. Solid oxide fuel cells (SOFC) constitute a promising alternative for power generation since they convert the chemical energy electrochemically directly to electricity. Their high operating temperature shows potential for energy integration with energy intensive units (e.g. steam reforming reactors). Although energy integration is an essential tool for increased efficiencies, it leads to highly complex process schemes with rich dynamic behavior, which are challenging to control. Furthermore, the use of process intensification for increased energy efficiency imposes an additional control challenge. This dissertation identifies and proposes solutions on design, operational and control challenges of integrated systems for hydrogen production and power generation. Initially, a study on energy integrated SOFC systems is presented. Design alternatives are identified, control strategies are proposed for each alternative and their validity is evaluated under different operational scenarios. The operational range of the proposed control strategies is also analyzed. Next, thermal management of water gas shift membrane reactors, which are a typical application of process intensification, is considered. Design and operational objectives are identified and a control strategy is proposed employing advanced control algorithms. The performance of the proposed control strategy is evaluated and compared with classical control strategies. Finally SOFC systems for combined heat and power applications are considered. Multiple recycle loops are placed to increase design flexibility. Different operational objectives are identified and a nonlinear optimization problem is formulated. Optimal designs are obtained and their features are discussed and compared. The results of the dissertation provide a deeper understanding on the design, operational and control challenges of the above systems and can potentially guide further commercialization efforts. In addition to this, the results can be generalized and used for applications from the transportation and residential sector to large--scale power plants.
  • Thumbnail Image
    Item
    Design of Control Configurations for Complex Process Networks
    (2015-05) Heo, Seongmin
    Tight integration is the rule rather than the exception in chemical and energy plants. Despite the significant economic benefits which result from efficient utilization of energy/material resources, effective control of plants with such integration becomes challenging; the network-level dynamics emerging from process interconnections and the model complexity of such plants limit the effectiveness of decentralized control approaches traditionally followed in plant-wide control. The development of effective control methods for complex integrated plants is a challenging, open problem. This thesis proposes methods to develop effective control strategies for two classes of process networks. In the first part of the thesis, a class of process networks, in which slow network-level dynamics is induced by large rates of energy and/or material recycle, is considered. A graph theoretic algorithm is developed for such complex material integrated process networks to i) identify the material balance variables evolving in each time scale, and ii) design hierarchical control structures by classifying potential manipulated inputs and controlled outputs in each time scale. The application of a similar algorithm developed for energy integrated networks to representative chemical processes is also presented. The second part of the thesis focuses on generic process networks where tight integration is not necessarily reflected on a segregation of energy and/or material flows. A method is developed to systematically synthesize control configurations with favorable structural coupling, using relative degree as a measure of such coupling. Hierarchical clustering methods are employed to generate a hierarchy of control configurations ranging from fully decentralized ones to a fully centralized one. An agglomerative hierarchical clustering method is first developed, in which groups of inputs/outputs are merged successively to form fewer and larger groups that are strongly connected topologically. Then, a divisive hierarchical clustering method is developed, in which groups of inputs/outputs are decomposed recursively into smaller groups. The developed methods are applied to typical chemical process networks.
  • Thumbnail Image
    Item
    Graph Representation And Distributed Control Of Lumped And Distributed Parameter System Networks
    (2019-05) Moharir, Manjiri
    Chemical plants are complex, integrated networks of individual process systems. The process system dynamics along with the interconnections among them make the task of controlling chemical plants challenging. Distributed control is a promising approach towards achieving plant-wide control of tightly integrated networks. The identification of sparsely interacting sub-networks in a given chemical network is key towards achieving superior performance from the distributed control structure. To this end, community detection algorithms have been adopted to determine the optimal decompositions for chemical networks by maximization of modularity. These algorithms are based on equation graph representations of the network. For lumped parameter system (LPS) networks, such representations are standard. Since chemical networks usually comprise lumped as well as distributed parameter systems (DPSs), this thesis aims at incorporating within the framework described above, the variables and topology of DPSs, to develop a unified framework to obtain optimal network decompositions (control structures) for distributed control. To this end, an equation graph representation for a generic DPS and a parameter which captures the strength of structural interactions among its variables analogous to relative degree in LPSs are proposed. A relationship is established between the length of the input-output path in the equation graph and the structural interaction parameter, which enables the incorporation of DPSs variables in the graph based community detection algorithms. Also, since in chemical networks, often the measurement of the entire state is not available and estimation of the unmeasured variables is a computationally expensive task, this thesis also addresses the problem of combined distributed state estimation and distributed control, using community detection for determining network decompositions for estimation as well as control.
  • Thumbnail Image
    Item
    Methods Of Distributed And Nonlinear Process Control: Structuredness, Optimality And Intelligence
    (2020-05) Tang, Wentao
    Chemical processes are intrinsically nonlinear and often integrated into large-scale networks, which are difficult to control effectively. The traditional challenges faced by process control, as well as the modern vision of transitioning industries into a smart manufacturing paradigm, requires the instillation of new perspectives and application of new methods to the control of chemical processes. The goal is to realize highly automated, efficient, well-performing and flexible control strategies for nonlinear, interconnected and uncertain systems. Motivated by this, in this thesis, the following three important aspects (objectives) for contemporary process control -- structuredness, optimality, and intelligence -- are discussed in the corresponding three parts. 1. For the control of process networks in a structured and distributed manner, a network-theoretic perspective is introduced, which suggests to find a decomposition of the problem according to the block structures in the network. Such a perspective is examined by sparse optimal control of Laplacian network dynamics. Community detection-based methods are proposed for input--output bipartite and variable-constraint network representations and applied to a benchmark chemical process. 2. For the optimality of control, we first derive a computationally efficient algorithm for nonconvex constrained distributed optimization with theoretically provable convergence properties -- ELLADA, which is applied to distributed nonlinear model predictive control of a benchmark process system. We derive bilevel optimization formulations for the Lyapunov stability analysis of nonlinear systems, and stochastic optimization for optimally designing the Lyapunov function, which can be further integrated with the optimal process design problem. 3. Towards a more intelligent diagram of process control, we first investigate an advantageous Lie-Sobolev nonlinear system identification scheme and its effect on nonlinear model-based control. For model-free data-driven control, we discuss a distributed implementation of the adaptive dynamic programming idea. For chemical processes where states are mostly unmeasurable, dissipativity learning control (DLC) is proposed as a suitable framework of input--output data-driven control, and applied to several nonlinear processes. Its theoretical foundations are also discussed.
  • Thumbnail Image
    Item
    Understanding the performance of decision strategies in dynamic environments.
    (2012-08) Meyer, Georg
    A decision strategy is systematic way of choosing among alternatives or eliminating options in order to arrive at a goal. Individuals apply decision strategies in dynamic environments that require repeated decision making where decisions are path-dependent, time-constrained, and the environment changes not only in response to the actions taken by the decision maker but also autonomously. In addition to being used by individual agents, decision strategies are found in organizations in the form of policies, guidelines, and algorithms. This research consists of three studies that apply a process control perspective to dynamic decision making. Study 1 investigates the features of decision strategies that affect performance. It finds that strategies perform well if they possess a strong mental model that accurately represents the decision problem or if they are well adapted to the problem environment. Based on these findings, Study 2 develops a machine learning approach to improve the mental model, and Study 3 develops an evolutionary approach to adapt decision strategies to a given environment. Both approaches are shown to be effective for constructing strategies with greater performance.