The present work introduces novel methodologies for damage detection and health monitoring of structural and mechanical systems. The new approach uses the angular velocity inside different mathematical forms, via a gyroscope, to detect, locate, and relatively quantify damage. This new approach has been shown to outperform the current state-of-the-art acceleration-based approach in detecting damage on structures. Additionally, the current approach has been shown to be less sensitive to environmental acoustic noises, which present major challenges to the acceleration-based approaches. Furthermore, the current approach has been demonstrated to work effectively on arch structures, which acceleration-based approaches have struggled to deal with. The efficacy of the new approach has been investigated through multiple forms of structural damage indices. The first methodology proposed a damage index that is based on the changes in the second spatial derivative (curvature) of the power spectral density (PSD) of the angular velocity during vibration. The proposed method is based on the output motion only and does not require information about the input forces/motions. The PSD of the angular velocity signal at different locations on structural beams was used to identify the frequencies where the beams show large magnitude of angular velocity. The curvature of the PSD of the angular velocity at these peak frequencies was then calculated. A damage index is presented that measures the differences between the PSD curvature of the angular velocity of a damaged structure and an artificial healthy baseline structure. The second methodology proposed a damage index that is used to detect and locate damage on straight and curved beams. The approach introduces the transmissibility and coherence functions of the output angular velocity between two points on a structure where damage may occur to calculate a damage index as a metric of the changes in the dynamic integrity of the structure. The damage index considers limited-frequency bands of the transmissibility function at frequencies where the coherence is high. The efficacy of the proposed angular-velocity damage-detection approach as compared to the traditional linear-acceleration damage-detection approach was tested on straight and curved beams with different chord heights. Numerical results showed the effectiveness of the angular-velocity approach in detecting damage of multiple levels. It was observed that the magnitude of the damage index increased with the magnitude of damage, indicating the sensitivity of the proposed method to damage intensity. The results on straight and curved beams showed that the proposed approach is superior to the linear-acceleration-based approach, especially when dealing with curved beams with increasing chord heights. The experimental results showed that the damage index of the angular-velocity approach outweighed that of the acceleration approach by multiple levels in terms of detecting damage. A third methodology for health-monitoring and updating of structure supports, which resemble bridges’ bearings, is introduced in this work. The proposed method models the resistance of the supports as rotational springs and uses the transmissibility and coherence functions of the output response of the angular velocity in the neighborhood of the bearings to detect changes in the support conditions. The proposed methodology generates a health-monitoring index that evaluates the level of deterioration in the support and a support-updating scheme to update the stiffness resistance of the supports. Numerical and experimental examples using beams with different support conditions are introduced to demonstrate the effectiveness of the proposed method. The results show that the proposed method detected changes in the state of the bearings and successfully updated the changes in the stiffness of the supports.

Vibration-based structural health monitoring (SHM) seeks to detect, quantify, locate, and prognosticate damage by processing vibration signals measured while the structure is operational. The basic premise of vibration-based SHM is that damage will affect the stiffness, mass or energy dissipation properties of the structure and in turn alter its measured dynamic characteristics. In order to make SHM a practical technology it is necessary to perform damage assessment using only a minimum number of permanently installed sensors. Deducing damage at unmeasured regions of the structural domain requires solving an inverse problem that is underdetermined and(or) ill-conditioned. In addition, the effects of local damage on global vibration response may be overshadowed by the effects of modelling error, environmental changes, sensor noise, and unmeasured excitation. These theoretical and practical challenges render the damage identification inverse problem ill-posed, and in some cases unsolvable with conventional inverse methods. This dissertation proposes and tests a novel interpretation of the damage identification inverse problem. Since damage is inherently local and strictly reduces stiffness and(or) mass, the underdetermined inverse problem can be made uniquely solvable by either imposing sparsity or non-negativity on the solution space. The goal of this research is to leverage this concept in order to prove that damage identification can be performed in practical applications using significantly less measurements than conventional inverse methods require. This dissertation investigates two sparsity inducing methods, L1-norm optimization and the non-negative least squares, in their application to identifying damage from eigenvalues, a minimal sensor-based feature that results in an underdetermined inverse problem. This work presents necessary conditions for solution uniqueness and a method to quantify the bounds on the non-unique solution space. The proposed methods are investigated using a wide range of numerical simulations and validated using a four-story lab-scale frame and a full-scale 17 m long aluminum truss. The findings of this study suggest that leveraging the attributes of both L1-norm optimization and non-negative constrained least squares can provide significant improvement over their standalone applications and over other existing methods of damage detection.

Cette thèse a pour but didentifier de changements dans le comportement dynamique dun système mécanique par le développement des techniques didentification, de détection et de recalage de modèle. Un endommagement ou une activation de non - linéarité est considéré responsable du changement. Une bien connue classification pour la détection dendommagement dans la littérature est utilisée dans la thèse, qui définit quatre niveaux dans lordre croissant de complexité: - Niveau 1 : Détection dendommagement : inspection de la présence dendommagement dans la structure. - Niveau 2 : Localisation dendommagement : détermination de position de lendommagement. - Niveau 3 : Evaluation de la gravité de lendommagement. - Niveau 4 : Prévision de la durée restant de vie de la structure. Selon la classification décrite ci-dessus, le problème de diagnostic dans cette thèse est abordé pour les trois premiers niveaux, i.e. détection, localisation et évaluation. Lidentification dendommagements ou dactivation de non - linéarité est basée sur une comparaison entre un état actuel et létat de référence (en normal condition). La thèse est organisée comme suit : Chapitre 1 présente une étude bibliographique sur des méthodes dindentification modale et de détection. Cette partie décrit quelques caractéristiques principales de systèmes non - linéaires et également des défis que présente la non - linéarité. Les problèmes de localisation et dévaluation sont discutés ensuite. Chapitres 2, 3 et 4 se concentrent sur la détection de défaut, par exemple lactivation de non linéarité ou lapparition dendommagement par trois méthodes respectivement : la Transformée en ondelettes (Wavelet transform), la Séparation aveugle au second ordre (Second-Order Blind Identification) et lAnalyse en composantes principales à noyau (Kernel Principal component Analysis). Seuls des signaux de sortie sont utilisés pour le traitement. Les deux premières méthodes réalisent la surveillance structurale par un processus didentification modale, tandis que la dernière méthode exerce directement dans des espaces caractéristiques déterminés par une fonction de noyau choisie. La détection peut être réalisée au moyen du concept dangle entre sous-espaces ou basée sur des statistiques. La robustesse des méthodes est illustrée sur une structure de poutre avec une non -linéarité géométrique à un bout ; ce modèle a été étudié dans le cadre dun projet de recherche européenne COST F3. Des autres exemples sont également considérés, tels quune maquette davion avec différents niveaux dendommagement et deux applications industrielles avec le but de contrôle de qualité sur un set dappareils électro - mécaniques et sur des joints de soudure. Chapitre 5 vise à la localisation dendommagement basée sur lanalyse de sensibilité des résultats de lAnalyse en composantes principales dans le domaine fréquentiel. La localisation est exécutée par la comparaison des sensibilités de composantes principales entre létat de référence (saine) et un état endommagé. Seules les réponses mesurées, e.g. réponses en fréquence (FRFs) sont nécessaires pour cet objectif. Après lanalyse de sensibilité au Chapitre 5, Chapitre 6 sadresse à lévaluation de paramètres, par exemple estimation dendommagements. Pour ce but, une procédure de recalage de modèle est exécutée. Cette procédure demande de construire un modèle analytique de la structure. Lanalyse de sensibilité pour la localisation dendommagement est illustrée par des données numériques et expérimentales dans des systèmes de masse - ressort et dans des structures de poutre. Une réelle structure, i.e. le pont I-40 à New Mexico qui a été détruite en 1993 est aussi examinée. Enfin, des conclusions sont retirées basées sur le travail réalisé et quelques perspectives sont proposées pour la continuation de cette recherche.

Numerical model usage has substantially increased in many industries. It is the aerospace industry that numerical models play possibly the most important role for development of optimum design. However, numerical models need experimental verification. This experimental verification is used not only for validation, but also updating numerical model parameters. Verified and updated models are used to analyze a vast amount of cases that structure is anticipated to face in real life. In this thesis, structural finite element model updating of a utility helicopter fuselage was performed as a case study. Initially, experimental modal analyses were performed using modal shakers. Modal analysis of test results was carried out using LMS Test.lab software. At the same time, finite element analysis of the helicopter fuselage was performed by MSC.Patran &amp / Nastran software. v Initial updating was processed first for the whole helicopter fuselage then, tail of the helicopter was tried to be updated. Furthermore, a new method was proposed for the optimum node removal location for getting better Modal Assurance Criterion (MAC) matrix. This routine was tried on the helicopter case study and it showed better performance than the Coordinate Modal Assurance Criterion (coMAC) that is often used in such analyses.

The performance of Enhanced Oil Recovery (EOR) processes is adversely affected by the heterogeneous distribution of flow properties of the rock. The effects of heterogeneity are further highlighted when the mobility ratio between the displacing and the displaced fluids is unfavorable. Polymer flooding aims to mitigate this by controlling the mobility ratio resulting in an increase in the volumetric swept efficiency. However, the design of the polymer injection process has to take into account the uncertainty due to a limited knowledge of the heterogeneous properties of the reservoir. Numerical reservoir models equipped with the most updated, yet uncertain information about the reservoir should be employed to optimize the operational settings. Consequently, the optimal settings are uncertain and should be revised as the model is updated. In this report, a feedback-control scheme is proposed with a model updating step that conditions prior reservoir models to newly obtained dynamic data, and this followed by an optimization step that adjusts well control settings to maximize (or minimize) an objective function. An illustration of the implementation of the proposed closed-loop scheme is presented through an example where the rate settings of a well affected by water coning are adjusted as the reservoir models are updated. The revised control settings yield an increase in the final value of the objective function. Finally, a fast analog of a polymer flooding displacement that traces the movement of random particles from injectors to producers following probability rules that reflect the physics of the actual displacement is presented. The algorithm was calibrated against the full-physics simulation results from UTCHEM, the compositional chemical flow simulator developed at The University of Texas at Austin. This algorithm can be used for a rapid estimation of basic responses such as breakthrough time or recovery factor and to provide a simplified characterization the reservoir heterogeneity. This report is presented to fulfill the requirements to obtain the degree of Master of Science in Engineering under fast track option. It summarizes the research proposal presented for my doctorate studies that are currently ongoing. / text

Polymer flooding is economically successful in reservoirs where the water flood mobility ratio is high, and/or the reservoir heterogeneity is adverse, because of the improved sweep resulting from the mobility-controlled oil displacement. The performance of a polymer flood can be further improved if the process is dynamically controlled using updated reservoir models and a closed-loop production optimization scheme is implemented. However, the formulation of an optimal production strategy is based on uncertain production forecasts resulting from uncertainty in spatial representation of reservoir heterogeneity, geologic scenarios, inaccurate modeling, scaling, just to cite a few factors. Assessing the uncertainty in reservoir modeling and transferring it to uncertainty in production forecasts is crucial for efficiently controlling the process. This dissertation presents a feedback control framework that (1) assesses uncertainty in reservoir modeling and production forecasts, (2) updates the prior uncertainty in reservoir models by integrating continuously monitored production data, and (3) formulates optimal injection/production rates for the updated reservoir models. This approach focuses on assessing uncertainty in reservoir modeling and production forecasts originated mainly by uncertain geologic scenarios and spatial variations of reservoir properties (heterogeneity). This uncertainty is mapped in a metric space created by comparing multiple reservoir models and measuring differences in effective heterogeneity related to well connectivity and well responses characteristic of polymer flooding. Continuously monitored production data is used to refine the uncertainty map using a Bayesian inversion algorithm. In contrast to classical approach of history matching by model perturbation, a model selection problem is implemented where highly probable reservoir models are selected to represent the posterior uncertainty in production forecasts. The model selection procedure yields the posterior uncertainty associated with the reservoir model. The production optimization problem is solved using the posterior models and a proxy model of polymer flooding to rapidly evaluate the objective function and response surfaces to represent the relationship between well controls and an economic objective function. The value of the feedback control framework is demonstrated with two examples of polymer flooding where the economic performance was maximized. / text

Out-of-plane laterally loaded masonry wall panels are still much used in modem structures. However due to their anisotropic and highly composite nature, it is extremely difficult to understand their behaviour and to date there is no analytical method that is capable of accurately predicting the response of masonry panels to the applied loadings. This is one of the major obstacles in analysing and designing masonry structures. This research studied a new method that accurately predicts the response of laterally loaded masonry wall panels. In this dissertation, the method of using corrector factors developed by previous researchers was further studied using model updating and artificial intelligence (AI) techniques based on previous experimental results of full scale wall panels tested in the University of Plymouth. A specialised non-linear finite element analysis (FEA) program was used to implement the method developed in this study. The analytical response was compared with other experimental results from different laboratories. Initially, it was found that there was some obvious noise in the experimental load deflection data, which made comparison between FEA and the experimental results very difficult. The research therefore proposed a methodology for minimising the experimental noise based on 3D surface fitting and regression analyses applied to lateral deflection experimental data. The next step was the detailed study of corrector factors using the numerical model updating procedure. Corrector factors were determined for various zones within a masonry panel (the Base Panel) by minimising the discrepancy between the experimental load deflection data and those obtained from non-linear FE analysis. A detailed model updating procedure was studied including the model analysis, the objective function and the constraint function for the genetic algorithm (GA). A uniqueness study to corrector factors was also carried out. The following step was undertaken to analyse general masonry wall panels using the findings of this study. The concept of zone similarities proposed by previous researcher, which was based on the relative distance of each zone from similar boundaries, was used for applying correctors from the base panel to the new panel to be analysed. A modified cellular automata (CA) model was used to match the similar zones between the new panel and the base panel. The generality and robustness of this method was validated using a number of masonry wall panels tested by various organizations. These walls were single leaf masonry wall panels of clay bricks with different boundary types, dimensions, with and without openings. The main finding in this research are that the boundary effects have a major influence on the response of masonry panels subjected to lateral loading, improperly defined boundary conditions in FEA are the main source of error in the past numerical analysis. Using the corrector factors that are able to properly quantify the actual boundary effects and make appropriate revisions, more accurate analysis is achieved and the predicted response of masonry walls match with their experimental results very well.

京都大学 / 新制・課程博士 / 博士(工学) / 甲第23858号 / 工博第4945号 / 新制||工||1772(附属図書館) / 京都大学大学院工学研究科社会基盤工学専攻 / (主査)教授 KIM Chul-Woo, 教授 高橋 良和, 准教授 北根 安雄 / 学位規則第4条第1項該当 / Doctor of Philosophy (Engineering) / Kyoto University / DFAM

Este trabalho apresenta metodologias para a construção e a sintonia de estimadores de estados não-lineares visando aplicações práticas. O funcionamento de um estimador de estados não-linear está calcado em quatro etapas básicas: (a) sintonia; (b) predição; (c) atualização da matriz de covariância de estados; (d) filtragem e suavização dos estados. As principais contribuições deste trabalho para cada uma destas etapas podem ser resumidas como segue: (a) Sintonia. A sintonia adequada da matriz de covariância do ruído de processos é fundamental na aplicação dos estimadores de estado com modelos sujeitos a incertezas paramétricas e estruturais. Sendo assim, foi proposto um novo algoritmo para a sintonia desta matriz que considera dois novos métodos para a determinação da matriz de covariância dos parâmetros. Este algoritmo melhorou significativamente a precisão da estimação dos estados na presença dessas incertezas, com potencialidade para ser usado na atualização de modelos em linha em práticas industriais. (b) Predição. Uma das etapas mais importantes para a aplicação do estimador de estados é a formulação dos modelos usados. Desta forma, foi mostrado como a formulação do modelo a ser usada em um estimador de estados pode impactar na observabilidade do sistema e na sintonia das matrizes de covariância. Também são apresentadas as principais recomendações para formular um bom modelo. (c) Atualização da matriz de covariância dos estados. A robustez numérica das matrizes de covariância dos estados usadas em estimadores de estados sem e com restrições é ilustrada através de dois exemplos da engenharia química que apresentam multiplicidade de soluções. Mostrou-se que a melhor forma de atualizar os estados consiste na resolução de um problema de otimização sujeito a restrições onde as estimativas fisicamente inviáveis dos estados são evitadas. Este também preserva a gaussianidade dos ruídos evitando que estes sejam mal distribuídos. (d) Filtragem e suavização dos estados. Entre as formulações estudadas, observou-se também que a melhor relação entre a acuracidade das estimativas e a viabilidade de aplicação prática é obtida com a formulação do filtro de Kalman estendido sujeita a restrições (denominada Constrained Extended Kalman Filter - CEKF), uma vez que esta demanda menor esforço computacional que a estimação de horizonte móvel, apresentando um desempenho comparável exceto no caso de estimativas ruins da condição inicial dos estados. Como uma solução alternativa eficiente para a estimação de horizonte móvel neste último caso, foi proposto um novo estimador baseado na inclusão de uma estratégia de suavização na formulação do CEKF, referenciado como CEKF & Smoother (CEKF&S). / This work presents approaches to building and tuning nonlinear state estimators aiming practical applications. The implementation of a nonlinear state estimator is supported by four basic steps: (a) tuning; (b) forecast; (c) state covariance matrix update; (d) states filtering and smoothing. The main contributions of this work for each one of these stages can be summarized as follows: (a) Tuning. An appropriate choice of the process-noise covariance matrix is crucial in applying state estimators with models subjected to parametric and structural uncertainties. Thus, a new process-noise covariance matrix tuning algorithm is presented in this work which incorporates two new methods for the parameter covariance matrix computation. The algorithm has improved significantly the state estimation accuracy when the presence of such uncertainties, with potential to be applied in on-line model update in industrial practice. (b) Forecast. One of the most important stages in applying state estimators is the used model formulation. In this way, it has been shown that the model formulation to be used in state estimator can impact on the system observability and noisecovariance matrices tuning. In this work it is also presented the main recommendations to formulate an appropriated model. (c) State covariance matrix update. The numerical robustness of the state covariance matrices used in unconstrained and constrained state estimators is illustrated by two chemical engineering examples tending to multiple solutions. It has been shown that the best technique to update the states consists in solving an optimization problem subjected to constraints, since it prevents from physically unfeasible states. It also preserves the noise gaussianity preventing from bad noise distribution. (d) States filtering and smoothing. Among the studied formulations, it was also noticed that the better relationship between performance and practical application is obtained with an extended Kalman filter formulation subjected to constraints (called Constrained Extended Kalman Filter - CEKF) because it requires small computational effort than MHE with comparable performance, except in case of poor guesses of the initial state. As an efficient solution for moving horizon estimation in the last case, it was proposed a new estimator based on the addition of a smoother strategy into the CEKF formulation, referred as CEKF & Smoother (CEKF&S).

Este trabalho apresenta metodologias para a construção e a sintonia de estimadores de estados não-lineares visando aplicações práticas. O funcionamento de um estimador de estados não-linear está calcado em quatro etapas básicas: (a) sintonia; (b) predição; (c) atualização da matriz de covariância de estados; (d) filtragem e suavização dos estados. As principais contribuições deste trabalho para cada uma destas etapas podem ser resumidas como segue: (a) Sintonia. A sintonia adequada da matriz de covariância do ruído de processos é fundamental na aplicação dos estimadores de estado com modelos sujeitos a incertezas paramétricas e estruturais. Sendo assim, foi proposto um novo algoritmo para a sintonia desta matriz que considera dois novos métodos para a determinação da matriz de covariância dos parâmetros. Este algoritmo melhorou significativamente a precisão da estimação dos estados na presença dessas incertezas, com potencialidade para ser usado na atualização de modelos em linha em práticas industriais. (b) Predição. Uma das etapas mais importantes para a aplicação do estimador de estados é a formulação dos modelos usados. Desta forma, foi mostrado como a formulação do modelo a ser usada em um estimador de estados pode impactar na observabilidade do sistema e na sintonia das matrizes de covariância. Também são apresentadas as principais recomendações para formular um bom modelo. (c) Atualização da matriz de covariância dos estados. A robustez numérica das matrizes de covariância dos estados usadas em estimadores de estados sem e com restrições é ilustrada através de dois exemplos da engenharia química que apresentam multiplicidade de soluções. Mostrou-se que a melhor forma de atualizar os estados consiste na resolução de um problema de otimização sujeito a restrições onde as estimativas fisicamente inviáveis dos estados são evitadas. Este também preserva a gaussianidade dos ruídos evitando que estes sejam mal distribuídos. (d) Filtragem e suavização dos estados. Entre as formulações estudadas, observou-se também que a melhor relação entre a acuracidade das estimativas e a viabilidade de aplicação prática é obtida com a formulação do filtro de Kalman estendido sujeita a restrições (denominada Constrained Extended Kalman Filter - CEKF), uma vez que esta demanda menor esforço computacional que a estimação de horizonte móvel, apresentando um desempenho comparável exceto no caso de estimativas ruins da condição inicial dos estados. Como uma solução alternativa eficiente para a estimação de horizonte móvel neste último caso, foi proposto um novo estimador baseado na inclusão de uma estratégia de suavização na formulação do CEKF, referenciado como CEKF & Smoother (CEKF&S). / This work presents approaches to building and tuning nonlinear state estimators aiming practical applications. The implementation of a nonlinear state estimator is supported by four basic steps: (a) tuning; (b) forecast; (c) state covariance matrix update; (d) states filtering and smoothing. The main contributions of this work for each one of these stages can be summarized as follows: (a) Tuning. An appropriate choice of the process-noise covariance matrix is crucial in applying state estimators with models subjected to parametric and structural uncertainties. Thus, a new process-noise covariance matrix tuning algorithm is presented in this work which incorporates two new methods for the parameter covariance matrix computation. The algorithm has improved significantly the state estimation accuracy when the presence of such uncertainties, with potential to be applied in on-line model update in industrial practice. (b) Forecast. One of the most important stages in applying state estimators is the used model formulation. In this way, it has been shown that the model formulation to be used in state estimator can impact on the system observability and noisecovariance matrices tuning. In this work it is also presented the main recommendations to formulate an appropriated model. (c) State covariance matrix update. The numerical robustness of the state covariance matrices used in unconstrained and constrained state estimators is illustrated by two chemical engineering examples tending to multiple solutions. It has been shown that the best technique to update the states consists in solving an optimization problem subjected to constraints, since it prevents from physically unfeasible states. It also preserves the noise gaussianity preventing from bad noise distribution. (d) States filtering and smoothing. Among the studied formulations, it was also noticed that the better relationship between performance and practical application is obtained with an extended Kalman filter formulation subjected to constraints (called Constrained Extended Kalman Filter - CEKF) because it requires small computational effort than MHE with comparable performance, except in case of poor guesses of the initial state. As an efficient solution for moving horizon estimation in the last case, it was proposed a new estimator based on the addition of a smoother strategy into the CEKF formulation, referred as CEKF & Smoother (CEKF&S).