Show simple item record

dc.rights.licenseRestricted to current Rensselaer faculty, staff and students. Access inquiries may be directed to the Rensselaer Libraries.
dc.contributorPequito, Sergio
dc.contributorPazour, Jennifer A.
dc.contributorSharkey, Thomas C.
dc.contributor.authorRomero, Orlando
dc.date.accessioned2021-11-03T09:21:32Z
dc.date.available2021-11-03T09:21:32Z
dc.date.created2021-01-21T12:37:49Z
dc.date.issued2020-08
dc.identifier.urihttps://hdl.handle.net/20.500.13015/2599
dc.descriptionAugust 2020
dc.descriptionSchool of Engineering
dc.description.abstractIn this work, we exploit tools from submodular optimization to help close the gap between the empirical and predicted performance of the greedy algorithm for problems in sensor and actuator selection. In particular, we provide novel tighter performance guarantees for the greedy algorithm on certain sensor selection formulations (regarding 1-step state estimation).
dc.description.abstractThis work explores the class of problems in the optimal design of (possibly large-scale) dynamical systems henceforth known as (optimal) sensor and actuator selection. In particular, it aims to create the mathematical tools towards a framework to systematically address this large class of problems, or important subclasses therein. More precisely, first we consider some time-varying process (engineered or otherwise) assumed to adhere to some structure that allows mathematical models to properly describe the system's evolution. Such a time-varying process constitutes what will be formalized as a dynamical system. The problems we explore are then posed as how to select, place, or schedule sensors and actuators in the system in a way that optimizes some performance criterion subject to pertinent selection constraints.
dc.description.abstractAt its core, sensors and actuators are any devices that serve to supervise and regulate the evolution of a dynamical system by inferring about and manipulating the system's internal state. Subsequently, the problems we explore fundamentally boil down to the design of optimal answers to the questions (for the underlying system at hand) "what should we be measuring?" and "what should we be manipulating?".
dc.description.abstractThe performance measures for the problem of optimal sensor and actuator selection typically reflect on the controllability, observability, or stability of the system, or on the system's estimation and prediction capabilities. Alternatively, we can consider the problem of minimizing the deployment cost among the selections that guarantee a certain performance threshold. The selection constraints, on the other hand, are usually based on physical constraints on the normal functioning of the sensors and actuators, or on natural limitations regarding a limited budget for the deployment costs on the sensors and actuators. For instance, the energy used by the sensors and actuators (e.g., battery and fuel consumption) may need to be limited, or the proper communication between mobile agents with sensing and actuating capabilities (e.g., swarm robotics and mobile sensor networks) may need to be guaranteed.
dc.description.abstractMathematically, optimal sensor and actuator selection translates to a very rich and heterogeneous class of optimization problems. Part of the motivation of this work is to leverage tools from operations research to address these control-theoretic problems. Furthermore, optimal sensor and actuator selection problems have, so far, been largely addressed on an individual basis through a variety of methods despite the strong conceptual similarity between the problems. From a more pragmatic, user-based perspective, this poses an unattractive scenario. First, it can be very time-consuming to determine which method would be appropriate from those available in the literature. Further, a small change in the original conditions on the underlying physical problem could result in a slightly different problem formulation, which could hinder the efficiency of a previously successful method or even make it inapplicable altogether. This lack of robustness could be resolved by the intended systematic methodology of this work.
dc.description.abstractThere is a substantial subclass of problems in optimal sensor and actuator selection that can be cast as combinatorial optimization problems. These are often extremely challenging to solve, from a computational perspective (i.e., they are often NP-hard). When an efficient solution cannot seem to be devised, we can sometimes derive efficient approximation algorithms or even computationally less demanding problems that serve as proxies for the original difficult problems. The standard greedy algorithm is one of such computationally efficient heuristics, and it has seen remarkable success as an approximation algorithm for various difficult combinatorial problems.
dc.description.abstractFurthermore, we propose a problem formulation for optimal actuator placement under symmetric structural controllability. We then formally prove that this proposed problem can be solved in polynomial time by implicitly constructing a polynomial-time algorithm together with proofs of correctness and an analysis of the computational complexity. Lastly, we discuss applications for sensor and actuator placement in the context of brain dynamics, and several numerical experiments illustrate our main results.
dc.language.isoENG
dc.publisherRensselaer Polytechnic Institute, Troy, NY
dc.relation.ispartofRensselaer Theses and Dissertations Online Collection
dc.subjectIndustrial and management engineering
dc.titleOptimal sensor and actuator selection : algorithms and performance guarantees
dc.typeElectronic thesis
dc.typeThesis
dc.digitool.pid180291
dc.digitool.pid180292
dc.digitool.pid180293
dc.rights.holderThis electronic version is a licensed copy owned by Rensselaer Polytechnic Institute, Troy, NY. Copyright of original work retained by author.
dc.description.degreeMEng
dc.relation.departmentDept. of Industrial and Systems Engineering


Files in this item

Thumbnail

This item appears in the following Collection(s)

Show simple item record