This thesis focuses on developing a high data rate wireless sensor network framework that could be integrated with hardware prototypes to monitor structural health of buildings. In order to better understand the wireless sensor network architecture and its consideration in structural health monitoring, a detailed literature review on wireless sensor networks has been carried out. Through research, it was found that there are numerous simulation software packages available for wireless sensor network simulation. One suitable software was selected for modelling the framework. Research showed that Matlab/Simulink was the most suitable environment, and as a result, a wireless sensor network framework was designed in Matlab/Simulink. Further, the thesis illustrates modeling of a simple accelerometer sensor, such as those used in wireless sensor networks in Matlab/Simulink using a mathematical description. Finally, the framework operation is demonstrated with 10 nodes, and data integrity is analyzed with cyclic redundancy check and transmission error rate calculations.

During strong earthquakes structural systems exhibit nonlinear behavior due to low-cycle fatigue, cracking, yielding and/or fracture of constituent elements. After a seismic event it is essential to assess the state of damage of structures and determine if they can safely resist aftershocks or future strong motions. The current practice in post-earthquake damage assessment relies mainly on visual inspections and local testing. These approaches are limited to the ability of inspectors to reach all potentially damaged locations, and are typically intended to detect damage near the outer surfaces of the structure leaving the possibility of hidden undetected damage. Some structures in seismic prone-regions are instrumented with an array of sensors that measure their acceleration at different locations. We operate under the premise that acceleration response measurements contain information about the state of damage of structures, and it is of interest to extract this information and use it in post-earthquake damage assessment and decision making strategies. The objective of this dissertation is to show that Bayesian filters can be successfully employed to estimate the nonlinear dynamic response of instrumented structural systems. The estimated response is subsequently used for structural damage diagnosis. Bayesian filters combine dynamic response measurements at limited spatial locations with a nonlinear dynamic model to estimate the response of stochastic dynamical systems at the model degrees-of-freedom. The application of five filters is investigated: the extended, unscented and ensemble Kalman filters, the particle filter and the model-based observer. The main contributions of this dissertation are summarized as follows: i) Development of a filtering-based mechanistic damage assessment framework; ii) Experimental validation of Bayesian filters in small and large-scale structures; iii) Uncertainty quantification and propagation of response and damage estimates computed using Bayesian filters.

Support vector machines (SVMs) are a set of supervised learning methods that have recently been applied for structural damage detection due to their ability to form an accurate boundary from a small amount of training data. During training, they require data from the undamaged and damaged structure. The unavailability of data from the damaged structure is a major challenge in such methods due to the irreversibility of damage. Recent methods create data for the damaged structure from finite element models. In this thesis we propose a new method to derive the dataset representing the damage structure from the dataset measured on the undamaged structure without using a detailed structural finite element model. The basic idea is to reduce the values of a copy of the data from the undamaged structure to create the data representing the damaged structure. The performance of the method in the presence of measurement noise, ambient base excitation, wind loading is investigated. We find that SVMs can be used to detect small amounts of damage in the structure in the presence of noise. The ability of the method to detect damage at different locations in a structure and the effect of measurement location on the sensitivity of the method has been investigated. An online structural health monitoring method has also been proposed to use the SVM boundary, trained on data measured from the damaged structure, as an indicator of the structural health condition.

Structural identification is the inverse problem of estimating the physical parameters, e.g. element masses and stiffnesses, of a model representing a structural system, using response measurements obtained from the actual structure subjected to operational or well-defined experimental excitations. It is one of the principal focal areas of modal testing and structural health monitoring, with the identified model finding a wide variety of applications, from obtaining reliable response predictions to timely detection of structural damage (location and severity) and consequent planning and validating of maintenance/retrofitting operations. However, incomplete instrumentation of the monitored system and ambient vibration testing generally result in spatially incomplete and arbitrarily normalized measured modal information, often making the inverse problem ill-conditioned and resulting in non-unique identification results. The problem of parameter identifiability addresses the question of whether or not a parameter set of interest can be identified from the available information. The identifiability of any parameter set of interest depends on the number and location of sensors on the monitored system. In this dissertation we study the identifiability of the mass and stiffness parameters of shear-type systems, including 3-dimensional laterally-torsionally coupled rigid floor systems, with incomplete instrumentation, simultaneous to the development of algorithms to identify the complete mass and stiffness matrices of such systems. Both input-output and output-only situations are considered, and mode shape expansion and mass normalization approaches are developed to obtain the complete mass normalized mode shape matrix, starting from the incomplete modal parameters identified using any suitable experimental or operational modal analysis technique. Methods are discussed to decide actuator/sensor locations on the structure which will ensure identifiability of the mass and stiffness parameters. Several possible minimal and near-minimal instrumentation set-ups are also identified. The minimal a priori information necessary in output-only situations is determined, and different scenario of available a priori information are considered. Additionally, tests for identifiability are discussed for both pre- and post-experiment applications. The different theoretical discussions are illustrated using numerical simulations and experimental data. It is shown that the proposed identification algorithms are able to obtain reliably accurate physical parameter estimates even under the constraints of minimal instrumentation, minimal a priori information, and unmeasured input. The different actuator/sensor placement rules and identifiability tests are useful for both experiment design purposes, to determine the necessary number and location of sensors, as well as in identifying possibilities of multiple solutions post-experiment. The parameter identification methods are applied for structural health monitoring using experimental data, and an approach is discussed for probabilistic characterization of structural damage location and severity. A perturbation based uncertainty propagation approach is also discussed for the identification of the distributions of mass and stiffness parameters, reflecting the variability in the test structure, using very limited measured and a priori information.

A Bayesian approach to system identification for structural control and health monitoring contains three main levels of inference, namely model assessment, joint state/parameter estimation and noise estimation. All of them have individually, or as a whole, been studied extensively for offline applications. In an online setting, the middle level of inference (joint state/parameter estimation) is performed using various algorithms such as the Kalman filter (KF), the extended Kalman filter (EKF), the Unscented Kalman filter (UKF), or particle filter (PF) methods. This problem has been explored in depth for structural dynamics. This dissertation focuses on the other two levels of inference, in particular on developing methods to perform them online, simultaneously to the joint state/parameter estimation. The quality of structural parameter estimates depends heavily on the choice of noise characteristics involved in the aforementioned online inference algorithms, hence the need for simultaneous online noise estimation. Model assessment, on the other hand, is an integral part of many engineering applications, since any analytical or numerical mathematical model used for predictive purposes is only an approximation of the real system. An online implementation of model assessment is valuable, amongst others, for structural control applications, and for identifying several models in parallel, some of which might be of deteriorating nature, thus generating some sort of alert. The performance of the proposed online techniques is evaluated using simulated and experimental data sets generated by nonlinear hysteretic systems. Upon completion of the study of hierarchical online system identification (diagnostic phase/estimation), a system/damage prognostic analysis (prognostic phase/prediction) is attempted using a gamma deterioration process. Prognostic analysis is still at a relatively early stage of development in the field of structural dynamics, but it can potentially provide useful insights regarding the lifetime of a dynamically excited structural system. The technique is evaluated on a data set recorded during an experiment involving a full-scale bridge pier under base excitation, tested to impending collapse.

This dissertation focuses on the development of data analytics approaches to two distinct important condition monitoring applications in civil infrastructure: structural health monitoring and road surface monitoring. In the first part, measured vibration responses of a major long-span bridge are used to identify its modal properties. Variations in natural frequencies over a daily cycle have been observed with measured data, which are probably due to environmental effects such as temperature and traffic. With a focus on understanding the relationships between natural frequencies and temperatures, a controlled simulation-based study is conducted with the use of a full-scale finite element (FE) model and four regression models. In addition to the temperature effect study, the identified modal properties and the FE model are used to explore both deterministic and probabilistic model updating approaches. In the deterministic approach (sensitivity-based model updating), the regularization technique is applied to deal with a trade-off between natural frequency and mode shape agreements. Specific nonlinear constraints on mode shape agreements are suggested here. Their capabilities to adjust mode shape agreements are validated with the FE model. To the best of the author's knowledge, the sensitivity-based clustering technique, which enables one to determine efficient updating parameters based on a sensitivity analysis, has not previously been applied to any civil structure. Therefore, this technique is adapted and applied to a full-scale bridge model for the first time to highlight its capability and robustness to select physically meaningful updating parameters based on the sensitivity of natural frequencies with respect to both mass and stiffness-related physical parameters. Efficient and physically meaningful updating parameters are determined by the sensitivity-based clustering technique, resulting in an updated model that has a better agreement with measured data sets. When it comes to the probabilistic approach, the application of Bayesian model updating to large-scale civil structures based on real data is very rare and challenging due to the high level of uncertainties associated with the complexity of a large-scale model and variations in natural frequencies and mode shapes identified from real measured data. In this dissertation, the full-scale FE model is updated via the Bayesian model updating framework in an effort to explore the applicability of Bayesian model updating to a more complex and realistic problem. Uncertainties of updating parameters, uncertainty reductions due to information provided by data sets, and uncertainty propagations to modal properties of the FE model are estimated based on generated posterior samples. In the second part of this dissertation, a new innovative framework is developed to collect pavement distress data via multiple vehicles. Vehicle vibration responses are used to detect isolated pavement distress and rough road surfaces. GPS positioning data are used to localize identified road conditions. A real-time local data logging algorithm is developed to increase the efficiency of data logging in each vehicle client. Supervised machine learning algorithms are implemented to classify measured dynamic responses into three categories. Since data are collected from multiple vehicles, the trajectory clustering algorithm is introduced to integrate various trajectories to provide a compact format of information about road surface conditions. The suggested framework is tested and evaluated in real road networks.

Advances in sensor technology and computer science in the last three decades have boosted the importance of system identification and vibration-based structural health monitoring (SHM) in civil infrastructure safety and integrity assessment. On the other hand, practical and financial issues in system instrumentation, maintenance, and operation have remained as fundamental problems obstructing the widespread use of SHM applications. For this reason, to reduce system costs and improve practicality as well as sustainability, researchers have been working on emerging methods such as wireless, distributed, mobile, remote, smart, multisensory, and heterogeneous sensing systems. Smartphones with built-in batteries, processor units, and a variety of sensors, have stood as a promising hardware and software environment that can be used as SHM components. Communication capabilities with the web, enable them to compose a smart and participatory sensor network of outnumbered individuals. Besides, crowdsourcing power offered by citizens, sets a decentralized and self-governing SHM framework which can even be pertained by very limited equipment and labor resources. Yet, citizen engagement in an SHM framework brings numerous challenges as well as opportunities. In a citizen-induced SHM scenario, the system administrators have limited or no control over the sensor instrumentation and the operation schedule, and the acquired data is subjected change depending on the measurement conditions. The citizen-induced errors can stem from spatial, temporal, and directional uncertainties since the sensor configuration relies on smartphone users’ decisions and actions. Moreover, the sensor-structure coupling may be unavailable where the smartphone is carried by the user, and as a consequence, the vibration features measured by smartphones can be modified due to the human biomechanical system. In addition, in contrast with the conventional high fidelity sensors, smartphone sensors are of limited quality and are subjected to high noise levels. This dissertation utilizes multisensory smartphone features to solve citizen-induced uncertainties and develops a smartphone-based SHM methodology which enables a cyber-physical system through mobile crowdsourcing. Using smartphone computational and communicational power, combined with a variety of embedded sensors such as accelerometer, gyroscope, magnetometer and camera, spatiotemporal and biomechanical citizen-induced uncertainties can be eliminated from the crowdsourced smartphone data, and eventually, structural vibrations collected from numerous buildings and bridges can be collected on a single cloud server. Therefore, unlike the conventional platforms designed and implemented for a particular structure, citizen-engaged and smartphone-based SHM can serve as intelligent, scalable, fully autonomous, cost-free, and durable cyber-physical systems drastically changing the forthcoming trends in civil infrastructure monitoring. In this dissertation, iOS is used as the application development platform to produce a smartphone-based SHM prototype, namely Citizen Sensors for SHM. In addition, a web-based software is developed and cloud services are implemented to connect individual smartphones to an administrator base and automate data submission and processing procedure accordingly. Finally, solutions to citizen-induced problems are provided through numerous laboratory and field test applications to prove the feasibility of smartphone-based SHM with real life examples. Through collaborative use of the software, principles and methodologies presented in this dissertation, smartphones can be the core component of futuristic smart, resilient, and sustainable city and infrastructure systems. And this study lays down an innovative and integrated foundation empowering citizens to achieve these goals.

Computer vision sensing techniques enable easy-to-install and remote non-contact monitoring of structures and have great potentials in field applications. This study will develop/implement novel computer vision techniques for two sensing systems for monitoring different aspects of infrastructures in challenging field conditions. The dissertation is therefore composed of two parts: robust measurement of global multi-point structural displacements, and accurate and robust monitoring of local surface displacements/strains. Computer vision based displacement measurement has become popular in the recent decade. The first part presents InnoVision, a vision sensing system developed to address a number of challenging problems associated with applying vision sensors to the measurement of multi-point structural displacement in field conditions that are rarely comprehensively studied in the literature. The challenging problems include tracking low-contrast natural targets on the structural surface, insufficient resolution for long distance measurement, inevitable camera vibration, and image distortion due to heat haze in hot weather. Several techniques are developed in InnoVision to tackle these challenges. Laboratory and field tests are conducted to evaluate the performance of these techniques. In the second part, another vision sensing system SurfaceVision is developed for accurate and robust monitoring two-dimensional (2D) structural surface displacements/strains. Important structures, such as nuclear power plants, need the continuous inspection of surface conditions. As an alternative to the human inspection, conventional digital-image-correlation (DIC) based methods have been applied to surfaces painted with speckle patterns in a controlled environment. However, it is highly challenging for DIC methods to accurately measure displacement on natural concrete surfaces in outdoor conditions with changing illumination and weather conditions. Additionally, common surface displacement measurement is based on segmenting the surface image into small subsets and tracking each subset individually through template matching, the surface displacement thus obtained has obvious discontinuity and low spatial resolution. Therefore, for applicability in the outdoor environment, SurfaceVision is proposed for accurate and robust monitoring of surface displacements/strains. Advanced computer vision techniques are developed/implemented to enable surface displacement measurement with high continuity, spatial resolution, accuracy, and robustness. An intuitive strain calculation method is also developed for converting surface displacements into surface strains. A numerical simulation is formulated based on four-point bending tests to validate the accuracy and robustness of SurfaceVision in surface displacements. Four-point bending experiments using reinforced concrete specimens are conducted to demonstrate the performance of SurfaceVision under different cases of optical noises and its effectiveness in predicting crack formations.

This dissertation describes the work conducted on error analysis for Brillouin Optical Time Domain Reflectometry (BOTDR), a distributed strain sensing technology used for monitoring the structural performance of infrastructures. Although BOTDR has been recently applied to many infrastructure monitoring applications, its measurement error has not yet been thoroughly investigated. The challenge to accurately monitor structures using BOTDR sensors lies in the fact that the measurement error is dependent on the noise and the spatial resolution of the sensor as well as the non-uniformity of the monitored infrastructure strain conditions. To improve the reliability of this technology, measurement errors (including precision error and systematic error) need to be carefully investigated through fundamental analysis, lab testing, numerical modelling, and real site monitoring verification. The relationship between measurement error and sensor characteristics is firstly studied experimentally and theoretically. In the lab, different types of sensing cables are compared with regard to their measurement errors. Influences of factors including fibre diameters, polarization and cable jacket on measurement error are characterized. Based on experimental characterization results, an optics model is constructed to simulate the Brillouin back scattering process. The basic principle behind this model is the convolution between the injected pulse and the intrinsic Brillouin spectrum. Using this model, parametric studies are conducted to theoretically investigate the impacts of noise, frequency step and spectrum bandwidth on final strain measurement error. The measurement precision and systematic error are then investigated numerically and experimentally. Measurement results of field sites with installed optical fibres displayed that a more complicated strain profile leads to a larger measurement error. Through extensive experimental and numerical verifications using a Brillouin Optical Time Domain Reflectometry (BOTDR), the dependence of precision error and systematic error on input strain were then characterized in the laboratory and the results indicated that a) the measurement precision error can be predicted using analyzer frequency resolution and the location determination error and b) the characteristics of the measurement systematic error can be described using the error to strain gradient curve. This is significant because for current data interpretation process, data quality is supposed to be constant along the fibre although the monitored strain for most of the site cases is non-uniformly distributed, which is verified in this thesis leading to a varying data quality. A novel data quality quantification method is therefore proposed as a function of the measured strain shape. Although BOTDR has been extensively applied in infrastructure monitoring in the past decade, their data interpretation has been proven to be nontrivial, due to the nature of field monitoring. Based on the measurement precision and systematic error characterization results, a novel data interpretation methodology is constructed using the regularization decomposing method, taking advantages of the measured data quality. Experimental results indicate that this algorithm can be applied to various strain shapes and levels, and the accuracy of the reconstructed strain can be greatly improved. The developed algorithm is finally applied to real site applications where BOTDR sensing cables were implemented in two load bearing piles to monitor the construction loading and ground heaving processes.

Orientador: Fábio Roberto Chavarette / Co-orientador: Mara Lúcia Martins Lopes / Banca: Francisco Villarreal Alvarado / Banca: Ivan Rizzo Guilherme / Resumo: Nesta dissertação de mestrado apresentam-se duas metodologias para o desenvolvimento de sistemas de monitoramento de integridade de estruturas mecânicas e aeronáuticas, utilizando técnicas de computação inteligente, tais como as redes neurais artificiais e os sistemas imunológicos artificiais. Neste contexto, emprega-se uma rede neural artificial ARTMAP-Fuzzy e o algoritmo de seleção negativa. Ambas as técnicas são empregadas para realizar a análise, identificação e caracterização das falhas estruturais decorrentes da estrutura. A principal aplicação destes métodos é auxiliar no processo de inspeção de estruturas mecânicas e aeronáuticas, visando detectar e caracterizar falhas, bem como, a tomada de decisões, a fim de evitar catástrofes/acidentes. Com estas propostas busca-se a concepção de novos sistemas de monitoramento de integridade estrutural que possam ser modificados facilmente, para atender a permanente evolução das tecnologias e da indústria. Para avaliar as metodologias propostas, foram realizados experimentos em laboratório para gerar um banco de dados de sinais capturados em uma viga de alumínio. Os resultados obtidos pelos métodos são excelentes, apresentando robustez e precisão / Abstract: In this dissertation presents two methodologies to develop health monitoring of aircraft structures and mechanical systems, using intelligent computing techniques such as artificial neural networks and artificial immune systems. In this context, uses an ARTMAP-Fuzzy artificial neural network and the negative selection algorithm. Both techniques are used for the analysis, identification and characterization of structural failure due to the structure. The main application of these methods is to assist in the inspection of mechanical and aeronautical structures, to detect and characterize flaws as well, making decisions in order to avoid disasters/accidents. With these proposals one seeks to designing new systems for structural health monitoring that can be modified easily to cater to permanent evolution technologies and industry. To evaluate the proposed methodologies, experiments were performed in the laboratory to generate a database of captured signals in an aluminum beam. The results obtained by the methods are excellent, with robustness and accuracy / Mestre