National Technical University of Athens (NTUA)
Συναρτήσεις φραγμένης κύμανσης και αντίστροφα προβλήματα
Abstract
dc:descriptionThe majority of the inverse techniques aiming at the identification of distributed physical parameters or the detection of buried objects of a geometrical nature (cracks, cavities, inclusions) embedded in the physical structures, distinguish the physical from the geometric reconstruction. Their implementation entails the understanding and analysis of the physical phenomenon, the possibility of measurements extraction, the formulation of the mathematical model describing the system and the development of the corresponding numerical reconstruction method. The requirements outlined above consist the most important steps in solving an inverse problem, i.e., determining the position and the shape of the inhomogeneity as well as identifying its physical characteristics. However, the investigation method and its effectiveness depend strongly on the a priori choice of the functional space in which the unknown system’s coefficients belong to. The present work adopts a new approach in an effort to bridge the traditional gap between physics and geometry. Namely, the functions of bounded variation (BV) are here recommended as the most appropriate functional setting hosting the solutions of inverse problems, exploiting their main intrinsic characteristic to represent simultaneously their discontinuity surfaces in the formation of their domain of definition. Accordingly, their investigation implies the identification of system’s both physical and geometrical features, providing an undivided inversion tool, which complies with the applications’ reality and allows us to look further into the physical interpretation of the phenomena.In the current work, emphasis is given to the problem of reconstructing inhomogeneities represented by discrete homogeneous bodies lying into finite domains, under the assumption that the unknown parameters of the model describing the inverse problem belong to the BV- space. Thesis’ context starts with solving the inverse problem of determining the conductivity function, α(x), of a conductive medium, from boundary measurements in the form of the Calderón operator, considering that α∈BV (Ch. 2). The designed herein methodology deals with a constructed minimization scheme implicating both the characteristics of the direct and inverse conductivity problem with the unknown BV- conductivity function, constituting an effective reconstruction scheme. Functional updating is implemented via a dual iterative scheme, where the related potential fields are located in the Sobolev space H^1, but the coefficient conductivity carries the BV- structure. This is accomplished by introducing an auxiliary function, ω(x), for accompanying α’s gradients acting as a mollifier of conductivity’s discontinuity surfaces. We adopt the dual approach “Half-Quadratic Minimization Approach”, encountered in bibliography as an image restoration technique, incorporating the variable ω into the regularization of the problem. This process defines a useful regularization type for the subsequent numerical implementation. In the realm of the half-quadratic technique, the variable ω is handling one at a time candidate function α and on the basis of this routine the minimization scheme is being optimized at every iteration so as to converge progressively to the sought interfaces. In general, the classical optimization techniques for defect identification exhibit local minima. To circumvent such pitfalls, we have taken further advantage of the dual variable ω by introducing into the minimization problem prior information about the possible location of the hidden discontinuity surfaces. The constructed minimization functional in conjunction with the aforementioned additional role of ω establish the presented inversion method. We validate and illustrate our theoretical results with numerical tests of some indicative two-dimensional conductivity profiles showing the efficiency of the method.The previous investigation has led to the invention of a novel reconstruction scheme, exploiting the duality of the half-quadratic approach and concurrently maintain the BV- structure of the coefficients α throughout the optimization process. Chapter 3 solves the inverse conductivity problem via this innovative reconstruction scheme. The denominated “Dual Self-Monitored TV- inversion” technique treats equivalently the fields α and ω, developing a monitoring mechanism controlling the dual variable ω, which simultaneously monitors the jump set of α. The developed numerical method permits the identification of a wider range of inhomogeneity profiles and at the same time with lower computational cost, as it converges in one algorithmic step to a local minimum (α,ω). The method’s performance is demonstrated on the simulation phantoms of Chapter 2 and additionally a more involved inclusion case has been investigated, with the numerical results revealing an impressively accurate reconstruction, even in the presence of noisy data.Next, the work is devoted to two types of inverse problems arising in the context of elasticity (Ch. 4). First, we study the inverse problem of identifying the Lamé moduli in a two-dimensional linear and isotropic elastic medium. We focus on the simultaneous recovery of both the Lamé coefficients λ,μ, considering them as bounded variation functions, α=(λ,μ)∈(BV)^2. The realization of this purpose invokes the concept of the dual monitoring mechanism implemented in Chapter 3 for the identification of a BV- conductivity. The suggested therein parameter identification approach has been extended to the vectorial case of the 2d- linearized elasticity, defining a generalization of the reconstruction scheme on the vector variable (α,ω). We employ the framework of the reconstruction of representantive elastic structures described by their Lamé parameters for the identification of a bimaterial. The resulting minimization scheme yields exact reconstruction solutions in accordance with the scalar conductivity case.Finally, we would like to create an analogous minimization pattern for the inverse problem of identifying a planar crack in a continuous isotropic and homogeneous elastic medium, derived from the hypothesis that the elastic field itself could support a BV- structure. We recast the dual variable ω(x) with the auxiliary variable r(x) and schedule a regularization term based on connecting efficiently the variable r along with the L^1- norm of the displacement fields, for the purpose of crack identification in an infinite elastic plane. We recur the inversion for different implementations varying with the length and the position of the crack generating a single pair of synthetic data each time. It turns out obviously that such an approach constitutes only a first stage in the implementation regime of this inverse problem, since the general problem is multiparametric. In practice, many variations of this problem type can be found in different physical contexts and are accordingly subject to several limitations depending on the practice area and the goal of the study. Nevertheless, the simulation results promote the applicability of the proposed inversion technique to this testing case.
Degree
thesis:*- Grantor dc:publisher
- National Technical University of Athens (NTUA)
- Year dc:date
- 2021
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Markaki, Vasiliki
- Μαρκάκη, Βασιλική
Subjects
dc:subject × 19- Αντίστροφα προβλήματα
- Συναρτήσεις BV
- Πρόβλημα Calderon
- Ομαλοποίηση φραγμένης κύμανσης
- Γραμμική ελαστικότητα
- Ανίχνευση ρωγμής
- Inverse problems
- BV functions
- Calderon problem
- Total variation regularization
- Crack identification
- Μαθηματικά
- Φυσικές Επιστήμες
- Εφαρμοσμένα μαθηματικά
- Μοντελοποίηση και Προσομοίωση
- Mathematics
- Natural Sciences
- Applied Mathematics
- Modeling and Simulation
Rights
- Language dc:language
- gre
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/50558