Ce travail de thèse porte sur l'étude et le développement de stratégies innovantes pour la résolution, basée sur l'utilisation de la simulation et de la théorie de l'apprentissage statistique, de problèmes inverses dans le domaine contrôle non destructif (CND) par méthodes électromagnétiques. L’approche générale adoptée consiste à estimer un ensemble des paramètres inconnus, constituant un sous-ensemble des paramètres décrivant le scénario de contrôle étudié. Dans les cas de CND, les trois applications classiquement visées sont la détection, la localisation et la caractérisation de défauts localisés dans le matériau inspecté. Ce travail concerne d’une part la localisation et la caractérisation des fissures et d’autre part l'estimation de certains paramètres de sonde difficiles à maîtriser ou inconnus. Dans la littérature, de nombreuses méthodes permettant de remonter aux paramètres inconnus ont été étudiées. Les approches d'optimisation standard sont basées sur la minimisation d'une fonction de coût, décrivant l'écart entre les mesures et les données simulées avec un solveur numérique. Les algorithmes les plus répandus se fondent sur des approches itératives déterministes ou stochastiques. Cette thèse considère le problème de l'estimation de paramètres inconnus dans une perspective d'apprentissage statistique/automatique. L’approche supervisée adoptée est connue sous le nom de d’apprentissage par l'exemple (LBE en anglais). Elle se compose d’une première phase, dite hors ligne, pendant laquelle un « modèle inverse » est construit sur la base de la connaissance d’un ensemble de couples entrée/sortie connu, appelé ensemble d’entraînement. Une fois la phase d’apprentissage terminée et le modèle généré, le modèle est utilisé dans une phase dite en ligne pour prédire des sorties inconnues (les paramètres d'intérêt) en fonction de nouvelles entrées (signaux CND mesurés appartenant à un second ensemble dit de test) en temps quasi-réel. Lorsqu’on considère des situations pratiques d'inspection, en raison du grand nombre de variables impliquées, la création d'un modèle précis et robuste n’est pas une tâche triviale (problème connu comme la malédiction de la dimensionnalité). Grâce à une étude approfondie et systématique, l’approche développée dans cette thèse a conduit à la mise en place de différentes solutions capables d’atteindre une bonne précision dans l’estimation des paramètres inversés tout en conservant de très bonnes performances en temps de calcul. Le schéma LBE proposé dans cette thèse a été testé avec succès sur un ensemble des cas réels, en utilisant à la fois des données synthétiques bruitées et des mesures expérimentales. / The research activity of the PhD thesis focuses on the study and development of innovative strategies for the solution of inverse problems arising in the field of Non-Destructive Testing and Evaluation (NDT-NDE), based on the use of statistical learning theory. Generally speaking, the objective of the optimization stage is the retrieval of the unknown parameters within the studied electromagnetic scenario. In the case of NDT-NDE, the optimization problem, in terms of parameters to estimate, is divided into three stages, namely detection, localization and characterization. This work mainly addresses localization and characterization of crack(s) and/or estimation of probe(s) parameters. Unknown parameters, constituting a subset of the parameters set describing the electromagnetic scenario, are robustly estimated using several approaches. Standard optimization approaches are based on the minimization, by means of iterative approaches like stochastic and/or deterministic algorithms, of a cost function describing the discrepancy between measurements and prediction. This thesis considers the estimation problem in a machine learning perspective, adopting well known Learning-By-Example (LBE) paradigm. In a so-called offline phase, a surrogate inverse model is first fitted on a set of known input/output couples, generated through numerical simulations. Then, in a so-called online phase, the model predicts unknown outputs (the parameters of interest) based on new inputs (measured NDT signals) in quasi-real time. When considering practical inspection situations, due to the large number of variables involved (known as curse of dimensionality), obtaining an accurate and robust model is not a trivial task. This thesis carries out a deep and systematic study of different strategies and solutions to achieve simultaneously good accuracy and computational time efficiency in the parameters estimation. Moreover, a particular emphasis is put on the different approaches adopted for mitigating the curse of dimensionality issue. The proposed LBE schema has been tested with success on a wide set of practical problems, using both synthetic noisy data and experimental measurements.

As a part of the Dutch development program Geo-impulse, which aims to half the occurrence of geotechnical failures in civil engineering projects inside the Netherlands by 2015, this master thesis is investigating how to trace imperfections in bored piles at an early stage. The objective is to carry out literature study on suitable methods and then focus the research on one particular method. The basis of the research is the single-hole sonic logging method. Field and laboratory measurements are carried out with the aim to investigate the detection range of the method as well as try to apply more advanced post-processing techniques. Results from the measurements are discussed and also a comparison between single-hole sonic logging and the better standardized test cross-hole sonic logging is made. The results indicate that single-hole sonic logging has a small detection range inside a bored pile, especially compared to cross-hole sonic logging. Also more advanced post-processing techniques fails or make the test to advance to use on a daily basis. Finally the recommendation is to carry on research with other techniques which in scientific papers have showed some promising results.

No description available.

No description available.

Master of Science / Department of Civil Engineering / Hani G. Melhem / The objective of this research is to use structure ambient random vibration response to detect damage level and location. The use of ambient vibration is advantageous because excitation is caused by service conditions such as normal vehicle traffic on a highway bridge, train passage on a railroad bridge, or wind loads on a tall building. This eliminates the need to apply a special impact or dynamic load, or interrupt traffic on a bridge in regular service. This research developed an approach in which free vibration of a structure is extracted from the response of this structure to a random excitation in the time domain (acceleration versus time) by averaging out the random component of the response. The result is the free vibration that includes all modes based on the sampling rate on time. Then this free vibration is transferred to the frequency domain using a Fast Fourier Transform (FFT). Variations in frequency response are a function of structural stiffness and member end-conditions. Such variations are used as a measure to identify the change in the structural dynamic properties, and ultimately detect damage. A physical model consisting of a 20 × 20 × 1670 -mm long steel square tube was used to validate this approach. The beam was tested under difference supports conditions varying from a single- to three-span continuous configuration. Random excitation was applied to the beam, and the dynamic response was measured by an accelerometer placed at various locations on the span. A numerical model was constructed in ABAQUS and the dynamic response was obtained from the finite element model subjected to similar excitation as in the physical model. Numerical results were correlated against results from the physical model, and comparison was made between the different span/support configurations. A subsequent step would be to induce damage that simulates loss of stiffness or cracking condition of the beam cross section, and that would be reflected as a change in the frequency and other dynamic properties of the structure. The approach achieved good results for a structure with a limited number of degrees of freedom. Further research is needed for structures with a larger number of degrees of freedom and structures with damage in symmetrical locations relative to the accelerometer position.

Sensitive nonlinear acoustic methods are suitable for material characterization. This thesis describes three nonlinear acoustic methods that are proven useful for detection of defects like cracks and delaminations in solids. They offer the possibility to use relatively low frequencies which is advantageous because attenuation and diffraction effects are smaller for low frequencies. Therefore large and multi-layered complete objects can be investigated in about one second. Sometimes the position of the damage is required. But it is in general difficult to limit the geometrical extent of low-frequency acoustic waves. A technique is presented that constrains the wave field to a localized trapped mode so that damage can be located.

Radiation damage occurs in reactor pressure vessel (RPV) steel, causing microstructural changes such as point defect clusters, interstitial loops, vacancy-solute clusters, and precipitates, that cause material embrittlement. Radiation damage is a crucial concern in the nuclear industry since many nuclear plants throughout the US are entering the first period of life extension and older plants are currently undergoing assessment of technical basis to operate beyond 60 years. The result of extended operation is that the RPV and other components will be exposed to higher levels of neutron radiation than they were originally designed to withstand. There is currently no nondestructive evaluation technique that can unambiguously assess the amount of radiation damage in RPV steels. Nonlinear ultrasound (NLU) is a nondestructive evaluation technique that is sensitive to microstructural features such as dislocations, precipitates, and their interactions in metallic materials. The physical effect monitored by NLU is the generation of higher harmonic frequencies in an initially monochromatic ultrasonic wave, arising from the interaction of the ultrasonic wave with microstructural features. This effect is quantified with the measurable acoustic nonlinearity parameter, beta. In this work, nonlinear ultrasound is used to characterize radiation damage in reactor pressure vessel steels over a range of fluence levels, irradiation temperatures, and material composition. Experimental results are presented and interpreted with newly developed analytical models that combine different irradiation-induced microstructural contributions to the acoustic nonlinearity parameter.

The thesis deals with the application of nondestructive testing methods (acoustic emission, electron microscopy, X-ray diffraction topography) to description and identification of stages in the fatigue process in aluminium alloys. The main contribution of this dissertation thesis consists of the approach and evaluation methodology of cyclic damage in the materials. It dealt especially with the verification of measured data by using data obtained using other nondestructive methods. The experimental results show that the field of nondestructive testing materials can fundamentally uncover the process of fatigue damage and improve current knowledge in this area. Combination of used methods has become a powerful tool for the identification and monitoring of the whole fatigue degradation process of materials.

La capacité à détecter des micro-défauts ou des défauts profonds dans les pièces métalliques constitue un enjeu important pour l'industrie de l'aéronautique ou du nucléaire. La technique de contrôle non destructif (CND) par courant de Foucault est souvent utilisée pour cette application. Cette thèse s’inscrit dans le cadre d'une collaboration ayant pour but la réalisation et l'intégration de micro-capteurs de champ magnétique basés sur l’effet de magnéto-impédance (MI) à des systèmes de détection par CND. Ces micro-capteurs de structure multicouche (ferromagnétique/conducteur/ferromagnétique) ont été élaborés en salle blanche par dépôt de films minces. Un traitement thermique sous champ magnétique a ensuite permis d’optimiser les propriétés du matériau et d’induire des anisotropies dans le plan des couches ferromagnétiques. Une méthode basée sur la double démodulation d’amplitude du signal de mesure a été proposée pour la caractérisation dynamique des capteurs. Les paramètres importants tel que la géométrie, l’anisotropie et la fréquence d’excitation ont été étudiés afin d’optimiser les caractéristiques. Les résultats ont montré la nécessité de polariser les capteurs en champ. Nous avons donc étudié la possibilité de réaliser, grâce à une technique de micromoulage épais, un microsolénoïde 3D et des travaux préliminaires sur l’intégration d'un capteur dans le microsolénoïde par transfert de film ont été effectués. Enfin, une étude théorique a été réalisée en tenant compte des résultats obtenus expérimentalement. Pour cela, le modèle de Landau-Lifshitz-Gilbert (LLG) a été implanté dans un code de calcul électromagnétique par éléments finis permettant de calculer l’impédance du capteur en fonction du champ magnétique appliqué. / The capability to detect micro-defects or buried flaws in the metallic parts is an important issue for the aerospace or nuclear industry. The technique of nondestructive testing (NDT) by eddy current is widely used for these applications. This thesis is part of collaboration project aimed at the realization and integration of magnetic field microsensors based on the magneto-impedance (MI) effect for the NDT detection systems. These multilayered structure microsensors (ferromagnetic/conductor/ferromagnetic) were realized in the clean room by thin film deposition method. A post-annealing step with magnetic field was then used to optimize the material properties and to induce magnetic anisotropy in the ferromagnetic layers. A method based on the double amplitude demodulation was proposed for the dynamic characterization of the sensors. The important parameters such as the geometry, the anisotropy and the driven frequency were studied in order to optimize the characteristics. The results showed that a bias field is necessary for the application. Therefore, we have investigated the possibility to realize, through thick micromoulding technique, a 3D microsolenoid and preliminary work on integrating a sensor in the microsolenoid by film transfer has been carried out. Finally, a theoretical study was investigated by taking into account the results obtained experimentally. For this purpose, the model of Landau-Lifshitz-Gilbert (LLG) has been implemented in an electromagnetic finite element calculation program in order to determine the impedance of the sensor as a function of the applied magnetic field.

Le contrôle non destructif (CND) par Courants de Foucault (CF) de défauts de fissuration est l’une des techniques les plus répandues dans de nombreux secteurs industriels. L’utilisation d’outils de modélisation adaptés permet d’améliorer les procédés de contrôle et la compréhension des données expérimentales observées. Ce travail de thèse, réalisé au CEA LIST et sous la direction de D. Lesselier (Laboratoire des Signaux et Systèmes), a pour objectif de proposer une approche de modélisation semi-analytique dédiée à la simulation du CND CF de défauts fins ou très fins dans une pièce plane conductrice composée de plusieurs couches. Il a fait l’objet d’une collaboration, dans le cadre du projet CIVAMONT, avec l’équipe Meander de l’University of Western Macedonia (Grèce), dirigée par le professeur T. Theodoulidis.Du point de vue de la simulation, la complexité du problème à traiter est liée aux particularités des défauts de fissuration : une ouverture très fine, un profil complexe et la possibilité d’avoir des ponts de conductivité entre les deux faces latérales du défaut. Ces caractéristiques expliquent la difficulté qu’ont les méthodes de simulation classiques, semi-analytiques ou purement numériques, à traiter efficacement ce type de configuration. Pour ces raisons, une approche dédiée aux défauts fins, fondée sur la méthode des éléments de frontière, a été développée. Elle présente l’avantage majeur de ne requérir qu’une discrétisation surfacique du défaut, en traitant analytiquement le calcul dans la direction de son ouverture. Après la résolution, avec la Méthode des Moments, de l’équation intégrale décrivant les interactions entre le champ d’excitation et le défaut, la réponse de la sonde est calculée en appliquant le théorème de réciprocité. Les développements théoriques réalisés dans cette thèse ont abouti à la mise en place d’une formulation générale permettant la prise en compte d’un nombre quelconque de défauts fins, d’orientations et des géométries différentes, pouvant être situés dans des couches différentes de la pièce. Par la suite, cette méthode innovante a été implémentée au sein de la plateforme de simulation CIVA, développée au CEA LIST, et a été validée expérimentalement à plusieurs reprises. Une extension de cette méthode a également permis la mise en place d’une approche la couplant à une modélisation volumique standard pour la simulation de configurations complexes comme le contrôle de fissures au voisinage d’un alésage. Ce travail, qui a fait l’objet d’une diffusion internationale affirmée, a permis de lever avec succès un certain nombre de difficultés théoriques et pratiques liées à la modélisation du CND CF de défauts fins. / Non Destructive Testing (NDT) with Eddy Current (EC) techniques are is widely employed in several industrial sectors for cracks detection. Numerical simulation tools are largely used in order to design sensors, understand the signals collected during the measurements process and to provide a support in expertise. This work has been accomplished inside CEA LIST in collaboration with L2S-Supélec. It is also a part of the CIVAMONT 2012 project, with the active participation of MEANDER laboratory members from University of Western Macedonia (Greece) and Technological Educational Institute of Western Macedonia (Greece). The main goal of our work has consisted in to developing a semi-analytical modeling approach, devoted to Eddy Current Testing (ECT) of multiple narrow cracks in planar multilayered structures. From the numerical point of view, simulation of multiple narrow cracks problems is a difficult task for classical methods, like for example the Volume Integral Method (VIM) or the Finite Element Method (FEM). The main issues reside in geometrical characteristics of narrow crack themselves. Indeed, a narrow crack presents a small opening as well as complex profile and a complex shape, with possible electrical contacts inside it. All these features increase enormously, with classical methods, the difficulty to simulate in rapid and/or precise way problems involving narrow cracks. We have tackled the narrow crack issue by developing a Boundary Element Method (BEM) dedicated to ECT signal modeling, starting from an approach presented in literature. Then, we have extended its capability to more realistic and challenging cases, such as the ECT of multilayered structures affected by complex narrow cracks. The principle of this method is to introduce additional assumptions, leading to the description of the crack perturbation as the effect of a dipole distribution, oriented toward the crack opening. Numerically speaking, such a description makes it possible to largely reduce, compared to the VIM, the number of unknowns that one needs to properly solve the problem. A particular attention has been devoted to the analytical formulation, in order to achieve generality, accuracy and efficiency. A precise derivation of the spectral-domain Dyadic Green Function (DGF) associated to our problem has first been developed. In this work, analytical expressions of the spectral-domain DGF have been obtained via the Discrete Complex Image Method (DCIM). Then, an accurate approximation of the spectral-domain DGF has been achieved via the Generalized Pencil of Function (GPOF) method. Therefore, the closed-form of the spectral-domain DGF, expressed under the form of Sommerfeld Integrals (SIs), has been calculated analytically. Finally, the integral equation(s) associate to the electromagnetic problem is solved by applying the Method of Moments (MoM).Validations with respect to experiments and commercial simulation software have been performed to test the model. A large set of configurations have been chosen in order to address realistic configurations involving multiple narrow cracks embedded in different layers of a given multilayered structure. The model proposed has shown its promising performance in terms of computational time compared with the VIM and the FEM. Moreover, a very good agreement with respect to the experimental data has always been observed. In the last and very recent part of our work, a coupled approach between BEM and VIM has been studied and developed in order to address, in a efficient way, problems where narrow cracks appear in the vicinity of with volumetric flaws (for example the simulation of fastener sites inspections). Comparisons with experimental measures have shown that the coupled approach is capable to achieve, overall, better results than the VIM and saves a lot of computational time.