Inverse problems aim at estimating the unknown excitations or properties of a physical system based on available measurements of the system response. For example, wave tomography is used in geophysics for seismic waveform inversion; in biomedical engineering, optical tomography is used to detect breast cancer tissue; in structural engineering, inversion techniques are used for health monitoring and damage detection in structural safety evaluation. Inverse solvers depend on the type of measurement data the unknown parameters to be estimated. The work in this thesis focuses on structural parameter identification based on static and dynamic measurements. As an integral part of the formulated inverse solver, the associated forward problem is studded and deeply investigated. In reality, the data are associated with uncertainties caused by measurement devices or unfriendly environmental conditions during data acquisition. Traditional approaches use probability theory and model uncertainties as random variables. This approach has its own limitation due to a prior assumption on the probability structure of uncertainty. This is usually too optimistic or not realistic. However, in practice, it is usually difficult to reliably assess the statistical nature of uncertainties. Instead, only bounds on the uncertain variables and some partial information about their probabilities are known. The main source of uncertainty is due to the accuracy of measuring devices; these are designed to operate within specific allowable tolerances, as defined by National Institute of Standards and Technology (NIST). Tolerances are performance requirements that fix the limit of allowable error or departure from true performance or value. Thus closed intervals are the most realistic way to model uncertainty in measurements. In this work, uncertainties in measurement data are modeled as interval variables bounded by their endpoints. It is proven that interval analysis provides guaranteed enclosure of the exact solution set regardless of the underlying nature of the associated uncertainties. This work presents a solution of inverse problems under measurements uncertainty within the framework of Interval Finite Element Methods (IFEM) and adjoint-based optimization techniques. The solution consists of a two-step algorithm: first, an estimate of the parameters is obtained by means of a deterministic iterative solver. Then, the algorithm switches to a full interval solution, using the previous deterministic estimate as an initial guess. In general, the solution of an inverse problem requires iterative solutions of the forward problem. Efficient and accurate interval forward solutions in static and dynamic domains have been developed. In particular, overestimation due to interval dependency has been drastically reduced using a new decomposition of the load, stiffness, and mass matrices. Further improvements in the available interval iterative solvers have been achieved. Conjugate gradient and Newton-Raphson methods to gether with an inexact line search are used in the newly formulated optimization procedure. Moreover Tikhonov regularization is used to improve the conditioning of the ill-posed inverse problem. The developed interval solution for the inverse problem under uncertainty has been tested in a wide range of applications in static and dynamic domains. By comparing current solutions with other available methods in the literature, it is proven that the developed method provides guaranteed sharp bounds on the exact solution sets at a low computational cost. In addition, it contains those solutions provided by probabilistic approaches regardless of the used probability distributions. In conclusion, the developed method provides a powerful tool for the analysis of structural inverse problem under uncertainty.

No description available.

We are interested in solving the time-harmonic inverse acoustic scattering problem for planar sound-soft obstacles. In this work, we introduce four methods for solving inverse scattering problems. The first method is a variation of the method introduced by Johansson and Sleeman. This method solves the inverse problem when we have the far field pattern given for only one incident wave. It is an iterative method based on a pair of integral equations used to obtain the far field pattern of a known single object. The method proposed in this thesis has a better computational performance than the method of Johansson and Sleeman. The second method we present is a multi-frequency method called the recursive linearization algorithm. This method solves the inverse problem when the far field pattern is given for multiple frequencies. The idea of this method is that from an initial guess, we solve the single frequency inverse problem for the lowest frequency. We use the result obtained as the initial guess to solve the problem for the next highest frequency. We repeat this process until we use the data from all frequencies. To solve the problem at each frequency, we use the first method proposed. To improve the quality of the reconstruction of the shadowed part of the object, we solve the inverse scattering problem of reconstructing an unknown sound-soft obstacle in the presence of known scatterers. We show that depending on the position of the scatterers, we may be able to obtain very accurate reconstructions of the entire unknown object. Next, we introduce a method for solving the inverse problem of reconstructing a convex sound-soft obstacle, given measures of the far field pattern at two frequencies that are not in the resonance region of the object. This method is based on the use of an approximation formula for the far field pattern using geometric optics. We are able to prove that for the reconstruction of the circle of radius $R$ and center at the origin, the size of the interval of convergence of this method is proportional to the inverse of the wavenumber. This procedure is effective at reconstructing the illuminated part of the object; however, it requires an initial guess close to the object for frequencies out of the resonance region. Finally, we propose a globalization technique to obtain a better initial guess to solve the inverse problem at frequencies out of the resonance region. In this technique, given the far field pattern of a convex object at two frequencies out of the resonance region, we use our extrapolation operator to generate synthetic data for low frequencies. We apply the recursive linearization algorithm, using as a single frequency solver the method that is based on geometric optics. We obtain an approximation of the object that can be used as the initial guess to apply the recursive linearization algorithm using the first method introduced as the single frequency solver.

Constitutive model parameters are identified during model calibration through trial-and-error process driven to fit test data. In this research, the calibration of constitutive models is formally handled as an inverse problem. The first phase of this research explores error propagation. Data errors, experimental biases (e.g. improper boundary conditions), and model errors affect the inversion of model parameters and ensuing numerical predictions. Drained and undrained tests are simulated to study the effect of these three classes of errors. Emphasis is placed on the analysis of error surfaces computed by successive forward simulations. The second phase of this research centers on test procedures. Conventional soil tests were developed to create uniform stress and strain fields; consequently, they provide limited amount of information, the inversion is ill-posed, and results enhance uncertainty and error propagation. This research examines soil testing using new, non-conventional loading and boundary conditions to create rich, diverse, non-uniform strain and stress fields. In particular, the flexural excitation of cylindrical soil specimens is shown to provide rich data leading to a more informative test than conventional geotechnical tests. The new test is numerically optimized. Then a set of unique experimental studies is conducted.

In this thesis, direct and inverse problems concerning nodal solutions associated with the one-dimensional p-Laplacian operators are studied. We first consider the eigenvalue problem on (0, 1), −(y0(p−1))0 + (p − 1)q(x)y(p−1) = (p − 1) £fw(x)y(p−1) (0.1) Here f(p−1) := |f|p−2f = |f|p−1 sgn f. This problem, though nonlinear and degenerate, behaves very similar to the classical Sturm-Liouville problem, which is the special case p = 2. The spectrum {£fk} of the problem coupled with linear separated boundary conditions are discrete and the eigenfunction yn corresponding to£fn has exactly n−1 zeros in (0, 1). Using a Pr¡Lufer-type substitution and properties of the generalized sine function, Sp(x), we solve the reconstruction and stablity issues of the inverse nodal problems for Dirichlet boundary conditions, as well as periodic/antiperiodic boundary conditions whenever w(x) £f 1. Corresponding Ambarzumyan problems are also solved. We also study an associated boundary value problem with a nonlinear nonhomogeneous term (p−1)w(x) f(y(x)) on the right hand side of (0.1), where w is continuously differentiable and positive, q is continuously differentiable and f is positive and Lipschitz continuous on R+, and odd on R such that f0 := lim y!0+ f(y) yp−1 , f1 := lim y!1 f(y) yp−1 . are not equal. We extend Kong¡¦s results for p = 2 to general p > 1, which states that whenever an eigenvalue _n 2 (f0, f1) or (f1, f0), there exists a nodal solution un having exactly n − 1 zeros in (0, 1), for the above nonhomogeneous equation equipped with any linear separated boundary conditions. Although it is known that there are indeed some differences, Our results show that the one-dimensional p-Laplacian operator is still very similar to the Sturm-Liouville operator, in aspects involving Pr¡Lufer substitution techniques.

This dissertation focuses on inverse problems for partial differential equations with multiscale coefficients in which the goal is to determine the coefficients in the equation using solution data. Such problems pose a huge computational challenge, in particular when the coefficients are of multiscale form. When faced with balancing computational cost with accuracy, most approaches only deal with models of large scale behavior and, for example, account for microscopic processes by using effective or empirical equations of state on the continuum scale to simplify computations. Obtaining these models often results in the loss of the desired fine scale details. In this thesis we introduce ways to overcome this issue using a multiscale approach. The first part of the thesis establishes the close relation between computational grids in multiscale modeling and sampling strategies developed in information theory. The theory developed is based on the mathematical analysis of multiscale functions of the type that are studied in averaging and homogenization theory and in multiscale modeling. Typical examples are two-scale functions f (x, x/[epsilon]), (0 < [epsilon] ≪ 1) that are periodic in the second variable. We prove that under certain band limiting conditions these multiscale functions can be uniquely and stably recovered from nonuniform samples of optimal rate. In the second part, we present a new multiscale approach for inverse homogenization problems. We prove that in certain cases where the specific form of the multiscale coefficients is known a priori, imposing an additional constraint of a microscale parametrization results in a well-posed inverse problem. The mathematical analysis is based on homogenization theory for partial differential equations and classical theory of inverse problems. The numerical analysis involves the design of multiscale methods, such as the heterogeneous multiscale method (HMM). The use of HMM solvers for the forward model has unveiled theoretical and numerical results for microscale parameter recovery, including applications to inverse problems arising in exploration seismology and medical imaging. / text

Thesis (Ph. D.)--University of Washington, 1996. / Vita. Includes bibliographical references (leaves [58]-60).

Wang Yuliang. / Thesis (M.Phil.)--Chinese University of Hong Kong, 2006. / Includes bibliographical references (leaves 70-75). / Abstracts in English and Chinese. / Chapter 1 --- Introduction to Inverse Problems --- p.1 / Chapter 1.1 --- Typical Examples --- p.1 / Chapter 1.2 --- Major Properties --- p.3 / Chapter 1.3 --- Solution Methods --- p.4 / Chapter 1.4 --- Thesis Outline --- p.4 / Chapter 2 --- Review of the Theory --- p.6 / Chapter 2.1 --- Basic Concepts --- p.6 / Chapter 2.1.1 --- Ill-posedness --- p.6 / Chapter 2.1.2 --- Generalized Inverse --- p.7 / Chapter 2.1.3 --- Compact Operators and SVE --- p.8 / Chapter 2.2 --- Regularization Methods --- p.10 / Chapter 2.2.1 --- An Overview --- p.11 / Chapter 2.2.2 --- Convergence Rates --- p.12 / Chapter 2.2.3 --- Parameter Choice Rules --- p.15 / Chapter 2.2.4 --- Classical Regularization Methods --- p.18 / Chapter 3 --- Ill-posedenss of Typical Inverse Problems --- p.23 / Chapter 3.1 --- Integral Equations --- p.24 / Chapter 3.2 --- Inverse Source Problems --- p.26 / Chapter 3.3 --- Parameter Identification --- p.34 / Chapter 3.4 --- Backward Heat Conduction --- p.37 / Chapter 4 --- Error Estimates for Parameter Identification --- p.39 / Chapter 4.1 --- Overview of Numerical Methods --- p.40 / Chapter 4.2 --- Finite Element Spaces and Standard Estimates --- p.43 / Chapter 4.3 --- Output Least-square Methods --- p.43 / Chapter 4.4 --- Equation Error Methods --- p.50 / Chapter 4.5 --- Hybrid Methods --- p.50 / Chapter 5 --- Numerical Experiments --- p.52 / Chapter 5.1 --- Formulate the Linear Systems --- p.53 / Chapter 5.2 --- Test Problems and Observations --- p.55 / Bibliography --- p.70

Thesis (Ph. D.)--University of Washington, 1998. / Vita. Includes bibliographical references (leaves [86]-88).

No description available.