Structural elements under compression can fail due to instability. Such type of failure may present itself as flexural buckling, which is the sudden change of configuration of one state to another at a critical compressive load. Actual structure characteristics, such as cracks, real joints, sudden changes of flexural stiffness, bracing, and actual boundary conditions, make current analytical buckling methods inefficient. For this reason, most structures are evaluated using Finite Element (FE) software based on the Finite Element Method (FEM). However, such a numerical method may have a high computational cost when parametric studies need to be performed. The motivation for this thesis was to develop an analytical alternative for the buckling analysis of columns and beam-columns. The thesis analyses jointed Euler-Bernoulli columns and beam-columns with N discontinuities due to step-changes of flexural stiffness, and influence of open edge cracks, real joints, bracing, and actual support conditions. The approach considered utilizes the Heaviside function to obtain a single piecewise expression for the deflection, slope, moment, and shear of columns and beam-columns. The use of the Heaviside function, along with proposed closed-form expressions, reduce the number of unknown integration constants to only four. Main findings are listed as following: (i) Linear buckling analysis of a column only depends on solving the determinant of the 4x4 matrix of unknown coefficients, obtained by imposing four boundary conditions. (ii) Solving the four unknown constants of integration results in performing a geometrical non-linear static analysis of the beam-column, which perfectly agrees with results obtained through FE software. (iii) Closed-form expressions reduce the number of unknown constants of integrations to only four, regardless of the number of sections composing a column or beam-column. Concluding remarks and recommendations include: (i) Closed-form expressions accurately provide linear buckling loads for columns and geometrical non-linear expressions for beam-columns. (ii) Equations solved algebraically can be stored and used for future analysis, saving the computation time of rerunning the analysis for specific columns or beam-columns. Having the equations expressed algebraically allows for the ease of performing parametric analysis and exploring the performance of new designs. (iii) Computational cost (time needed for the computer to obtain the results) may depend on the mathematical software used for the calculations. Other software may result in having lower computational time for the same calculations.

Structural joints influence the design strength, material requirement of a structure. Structural joints experience damping dissipation due to friction damping or hysteresis damping. Damping is often used for reducing the vibrations in a structure. However, large amount of energy dissipation leads to deterioration of the material used for constructing the joint. Hence it is important to identify the system parameters like stiffness, viscous damping, friction force as well as the hysteretic restoring force that cause the energy dissipation in the structure. For identifying the uncertain system parameters like stiffness, viscous damping and magnitude of friction force, the SINDy algorithm is extended by using stick and slip temporal constraints. This is done by segregating the data of external forcing and response of SDoF system, applying the existing SINDy algorithm and applying the sticking and slipping conditions in the time domain. The proposed Extended SINDy approach estimates the system parameters more accurately compared to the existing SINDy algorithm. For studying the hysteresis in the structural joints, a pinned column base-plate was considered in an elastic region. Further, the Dahl model with different slope parameter for each branch of moment-rotation hysteresis is employed. The correct values of parameters are estimated using the Bayesian Optimization technique. This procedure yields a functional form representing a resisting hysteretic moment-rotation behaviour in a structural joint with good accuracy.

This research investigates how the current friction design standard should be improved, specifically for monopile sea fastening with polymers. Literature and results from tests performed previously to this research are investigated in order to obtain a complete overview of the factors that should be considered in the design standard. Additional tests are carried out on factors for which literature and previous test exercises are inconclusive. Friction test methods are compared to determine the best method for testing monopile sea fastening friction. The total test variability is analyzed to determine why safety measures are required. The validity of these measures is studied with a large set of test results. Finally, modelling as an alternative for testing is investigated.

Nautical radar systems are subjected to random vibrations during their service life, which may lead to the degradation of the materials and a possible fatigue failure. Assessing the impact of these vibrations to the structural integrity of the radar is crucial for drafting appropriate maintenance plans and therefore, the development of a new tool is required, which will be capable of evaluating the fatigue lifetime of the radar system. This thesis examines a 5.7-meter-long radar antenna, supported by a lattice tower with a height of 20 meters, located in Rijkswaterstaat’s test environment in Stellendam. Although the lattice structure and the radar system are coupled, they are treated separately in this study. Firstly, a Finite Element Model of the tower is developed in ANSYS, where the radar is represented by an added mass at the top floor. The aim of the model is to identify the vibrations of the structure under the influence of the incoming wind. The stochastic description of the turbulent wind is taken into account in the evaluation of the wind loads and a random vibration analysis is performed, resulting into the response of the structure in the frequency domain. Furthermore, a second, simplified Finite Element Model of the radar system is built in ANSYS. Time series of the fluctuating wind velocity are generated and the wind load acting on the radar is approximated by the aerodynamic force in the along-wind direction, considering also the rotation of the radar antenna. In addition, the previously obtained response of the lattice structure serves as input for the radar model, representing the vibrations induced to the radar due to the motion of the structure below. The output of the model is the occurring stress response, which is subsequently used to assess the fatigue damage of the radar antenna. The results show that the development of stresses at the bottom surface of the antenna mainly occurs at its central area. Given the assumptions used throughout the thesis, the motion of the lattice tower is found to be the governing loading condition for the resulting stresses and the magnitude of the motion significantly influences the fatigue damage of the antenna. Finally, the fatigue lifetime appears to be very sensitive to the construction materials used for the radar system. These results should be verified by future measurements in order to improve the suggested models.

The dynamic behavior of structural components can largely change in the presence of damages. Understanding this behavior is of particular importance for critical engineering systems, and in particular bridges. Damage identification methods forms a key objective in structural health monitoring of bridges so many researches have been conducted in this area. In this thesis, damage identification techniques on beam bridges under moving vehicle loads will be presented in order to produce useful conclusions about the assessment of existing bridges by investigating various numerical applications of different scenarios. The first objective of this thesis is to derive the analytical expressions needed to be able to predict the dynamic response of many different cases of bridges so that as many real scenarios as possible can be treated. This means that these expressions would be used to investigate damaged beam bridges that can be modelled as an assembly of beams with any number of different material properties, any type of interface or boundary conditions and any number of cracks. For this reason an approach to analyze the bridge as an assembly of n piecewise homogeneous damaged Euler-Bernoulli beams jointed at their edges, will be presented, using the generalized functions to obtain a single expression of the solution which depends on the 4 integration constants associated with the boundary conditions. The closed-form expressions of these 4 constants will be provided. Furthermore, in the presence of internal or externals springs, translational or rotational, additional constants representing the discontinuities have to be taken into account and are computed by considering one additional condition for each discontinuity. The feasibility of this approach and the corresponding analytical formulations is shown with two numerical applications that include all the different capabilities mentioned. Moreover, the implementation of these expressions in a deterministic approach for damage localization is presented, mainly as another example of the many possibilities of the use of analytical formulations instead of other approaches and as an introduction of the so called Inverse Problem with deterministic and probabilistic methods. The second objective concerns the optimization of damage identification on bridges by comparing different quantities that are evaluated while measuring the response of the bridge (direct monitoring) and the response of the moving vehicle when it passes along the bridge (indirect monitoring). First, the governing equations for the dynamic response of these models are derived, considering the crack(s) as a rotational spring, the bridge as an Euler-Bernoulli beam (or multiple with different properties) and the moving vehicle as a spring-mass system. In this manner, the dynamic response of the bridge is calculated (modal characteristics and displacement) as well as the one of the moving oscillator (displacement and acceleration) and the reaction force acting on the surface of the beam from the moving vehicles. Numerical applications with different beam properties and different number of cracks are performed, using MATLAB for the analytical expressions and SAP2000 for the finite element model, to derive the optimal quantity to be used for damage identification. Lastly, the results are validated by considering and comparing an alternative way of modelling crack, namely as a zone with reduced rigidity, for the same numerical examples, leading to the same conclusions about the crack(s) identification. Last but not least, the third objective of this thesis is to be able deal not only with the widely used time-invariant damages, namely the always-open crack model, but also with time-variant damages and in this case with the switching crack model. To achieve this, the analytical expressions for the closed-form solutions of the mode shapes derived for the always-open crack are modified to be able to tackle the switching crack model by introducing a Boolean switching crack array which identifies open cracks, modelled as rotational springs. These new expressions would still be able to be used for any number of Euler-Bernoulli beams, any type of interface or boundary conditions and any number of switching cracks. Then, the governing equations for the dynamic response of this model are derived, considering the moving vehicles as moving masses in order to validate the approach with numerical examples existing in the literature and then by introducing its new capabilities. Further, as the computational strategy has been validated, a comparison between time-variant and time-invariant damages is performed concerning crack identification, so that the reader would recognize the importance of understanding the dynamic behavior of different ways of modelling damage in complicate engineering systems like bridges.

Perforated monopiles show promise in providing a better alternative to the commonly used jacket-like substructures used in intermediate water depths in the range of 30 to 120 m. By introducing perforations near the vicinity of the splash zone the wave loads on the monopile can be mitigated and the fatigue damage reduced, which is the main advantage jackets have over monopiles. Furthermore, perforated monopiles would be easier to install and manufacture compared to jacket structures, which has the potential to reduce the Levelized Cost of Electricity (LCoE) considerably. Perforated monopiles also have the benefit of reduced (local) scour, reduced corrosion damage and providing a habitat for marine life. To optimize the design of perforated monopiles fast novel surrogate models are needed to determine the (wave) load reduction by the introduction of perforations on the monopile. As traditional (full) order Computational Fluid Dynamics (CFD) models are slow and expensive to evaluate. Using a surrogate model a wider range of geometries can be used in the early design optimization steps. Promising designs can then be further evaluated using full models. This thesis develops a surrogate model for the characterization of the (hydrodynamic) loads on perforated monopiles using Convolution Neural Networks (CNN). A dataset is created of 15,561 samples, which considers geometries of varying number of perforations, porosity and ‘angle of attacks’. A Finite Element Method (FEM) CFD model is made in Gridap for a ‘creeping flow’ to determine the flow fields. And determines a load Reduction Factor (RF) by the introduction of perforations. Furthermore, using a Signed Distance Function (SDF) a representation of the geometry is made for use in the CNN model. Using PyTorch a CNN model is used to predict the RF for a certain input. The model combines multiple convolutional layers with an optional regression head (with linear layers). Five different models are created. Three for representing the geometry by using the SDF, decomposing the SDF in its x and y distance components and the binary representation. Furthermore, two models combine the SDF or distance components with the flow fields from the CFD model. The results show that a speedup of 386 times is achieved by using the surrogate model, showing the main benefit of using a surrogate model. This is likely to improve considerably in further research when turbulence is taken into account. Furthermore, each of the five models show a good accuracy in predicting the RF. A general trend is observed that the more information is provided in the input to the CNN model the better the accuracy tends to be. The combination of implicit representation of the geometry by using distance components with the CFD flow fields shows the best accuracy with an expected maximum relative error rate of 0.94% within the parametric space of the dataset.

Wind turbines are typically designed for an operational life of 20-25 years. The operation of the assets can be extended beyond their design life if structural components have sufficient reserves left. One approach is to monitor fatigue loads and compare these with design assumptions to determine the remaining useful lifetime of the assets. A major challenge is that sensors for measuring the stress history, such as strain gauges, only deliver local information. Monitoring of every hot spot is technically and financially not feasible due to cost and access restrictions. In addition, strain gauges only have a limited lifetime when compared to accelerometers. Several response estimation or extrapolation methods have been proposed in literature to tackle this problem. All of them are Kalman filter based methods, with the exception of the Modal Decomposition and Expansion Method. A new Kalman filter based method has been recently proposed in literature called Gaussian Process Latent Force Model. The aim of this work is to assess this new method with respect to existing ones both theoretically and numerically. Theoretically, the Kalman filter based methods always rely on white gaussian noise assumptions for the unknown loads, and the modal decomposition and expansion method disregard measurement imperfections. The new method improves upon these assumptions by providing a flexible stochastic definition for the unknown load, and by taking into account measurement imperfections. The numerical analysis is restricted to a comparison with respect to the modal decomposition and expansion method given the different theoretical backgrounds. The comparison is realised upon a simulation which illustrates and validates the methods, and upon a real-life onshore wind turbine equipped with accelerometers and strain gauges. Measured strains are compared with estimated strains. The accuracy of each method is quantified using the mean absolute error and by the correlation between measurement and estimation. The higher accuracy obtained shows that the new method is an improvement upon existing methods. This is further extended in the calculation of Damage Equivalent Loads. The result shows a relative error that, depending on operational conditions, ranges within 20-40[%] for the modal decomposition and expansion method and less than 10[%] for the new method. These results show that the novel Gaussian Process Latent Force Model method should be taken into account for response estimation when accuracy is relevant. Future works should aim on developing a mechanical model that better capture the real behaviour of the wind turbine, as the accuracy of the response estimation methods is mainly controlled by the validity of the underlying assumptions of the mechanical model. Furthermore, the strain estimations should be sought in the whole frequency range, this can be realised by including measurements that deliver this information: GPS sensors or inclinometers for example.

As the average age of bridges in the Netherlands is increasing and loads are rising, more research on the capacity of these bridges is required. One of the methods to determine if a bridge still has sufficient safety is proof load testing. Stop criteria are values of structural responses at which further loading may course irreversibly damage to the structure. Measuring all currently proposed stop criteria requires a lot of single sensors, and it is holding back the application of the method in practice. A method to improve the applicability of proof load testing in practice is to simplify the sensor setup by using semi continuous fiber optical sensors. In this research, a fiber optical measurement system is developed to measure stop criteria for proof load testing. For application of the system to the slab a stiff glued anchorage system is designed, which is able to transfer the concrete strains to the reusable, long gauge fiber optic sensor system. The performance of this anchorage system is checked by preliminary experiments that include the comparison of various glues and investigates time-dependent effects. An installation method for these anchors is developed to improve practical applicability. Stop criteria are developed and checked by a 1:2 scale experiment on a concrete slab bridge. The results from the optical fiber sensors measure the global behavior of the slab bridge such as cracking and deflection well. Internal optical fiber sensors were included within the reinforcement. These sensors provide information on the strains up until the yielding point of the reinforcement bars. In conclusion, the fiber optical measurement system does provide insight on global and local behavior of the slab, and is able to monitor stop criteria. More importantly, the critical location cannot be missed with this sensor setup, as well as that more simplifications in the measurement setup are possible. It is therefore recommended to apply this measurement technique in future applications of proof load testing.

Given the desire to construct more structures using sustainable building materials, demand for wood and other bio-based building materials has risen dramatically over the decade. While timber itself is a carbon-neutral material and can sometimes even be carbon-negative, reusing wooden structural members made that have been in service for several years can widen the approach to structural design using wood. In the extensive network of rivers and canal systems that the Netherlands has, the banks are often covered with sheet-pile walls and a majority of these are made of timber. Tropical hardwood, especially Azobé (Lophira alata), due to its high biological resistance to decay, is used to make these sheet-piles. However, when exposed to the groundwater table for a long period of time, the wood undergoes decay due to bacteria destroying the cellulose slowly, while the lignin remains constant, and over decades the large cellulose molecules are replaced by water making the walls weaker. In this study, the characteristic mechanical properties of Azobé sheet-piles that have been in service for 57 years have been found, so that they can be assigned with an appropriate strength class and reused. Destructive, quasi-destructive and non-destructive tests have been performed on the sheet-pile boards to understand the correlation between them and to also ascertain to what level tests on timber specimens that do not affect their usability can be reliable. Since the knowledge of how visual grading can be performed on used, decayed structural timber, especially hardwood specimens is limited, a methodology has been developed in line with NEN-EN 14081-1:2019 along with the definition of a visual decay score. The results from the RPD tests are quantified in terms of the resistographic measure value to identify whether, in tropical hardwood, is there any effect in the drilling direction and whether this value can qualitatively or quantitatively describe the actual mechanical strength of the sheet-pile boards. The stress-wave tests and the four-point bending test are used to calculate the strength and stiffness of the boards, which determine the characteristic values, that are also based on their wet density (at which the tests are conducted). The results obtained are also analyzed for occurring patterns in terms of the location of the board in the sheet-pile wall (top or bottom), testing configuration (E-side or W-side-up) and variation in thickness within the boards due to decay. The bending test performed on the boards is modelled numerically in multiple iterations with curved-shell, layered-shell and 3D brick elements also varying the respective material models to find which one of them is best suitable to model bending of timber. The load-sharing mechanism observed when grouping multiple timber specimens has been simulated numerically to predict the characteristic load-sharing factor of the sheet-pile wall system.

In slender high-rise buildings wind-induced accelerations can become the governing design criterion. These issues can be solved by increasing mass, increasing stiffness, or increasing the damping ratio. The focus of this research is on the last option and investigates the opportunities of supplemental damping in the Dutch high-rise building context. The research report answers the following main research question: In what way can supplementary damping be applied in Dutch, slender, tall buildings to efficiently meet the structural design requirements? A literature research was conducted to the topics of wind-engineering, high-rise structures, and supplemental dampers. To answer the main research question two research methods have been applied. First, a variant study was performed in Karamba, a FEA plug-in for Rhino Grasshopper. Building height, floor mass, outrigger, stiffness limits, and damping ratio were varied. From this analysis charts were generated from which it could be determined which design requirement is governing. In this way opportunities for the application of supplemental damping could be determined. In the last phase of the research the mitigating effects of tuned mass damper (TMD) were modelled with the use of the theory of random vibrations and modal analysis. With a developed python script the reduced accelerations could be computed and the parameters of a TMD could be determined.

Damen is designing more and more slender hull forms, which leads to a small dock block contact area, and therefore a high load on the block bed. Dry docking is achieved by using multiple blocks located along the longitudinal length of the ship. Dock blocks are constituted by several layers, normally three or more. The configuration of the block bed (concrete in combination with hard­ and softwood) creates the support of the vessel with non­linear behaviour. This study focuses on a prediction of the blockload in the dry dock, taking into account non­linear material behaviour of the timber layers. Nowadays, the approach to predict the dock block load makes the assumption of linear elastic behaviour of the timber layers. The loads produced can cause a non­linear behaviour of each block, which might lead to an over­stress failure of one or more blocks. The assumption of linear elastic behaviour possibly results in wrong load distributions, which eventually leads to a redistribution of the loads on the other blocks. In turn this can force other block failures and ultimately this produces damages of the hull forms. The consequences of a poor dry docking analysis are potentially catastrophic. A docking failure can lead to extensive ship and dock damage, disruption of docking schedules, and loss of the ship to active duty until repairs can be made. Two set of models are conducted to predict the load on a dock block and evaluate the influence of the non­linear behaviour. The Timoshenko beam theory with multiple springs is used to calculate the load on every single dock block location. The load on every particular location is used in the model for a single dock block formed by elastically connected beams. A double Winkler foundation system represents the interaction of the stacked block layers and dictates the non­linear behaviour. Non­linear material properties are captured by the secant modulus of the top layer. A test campaign pointed out that for each layer of nominal identical specimens a large variability in material properties is noticed. Material properties found in the literature were different from those obtained during experiments. Moisture content has a significant influence on linear and non­linear effect, with the Young’s modulus and yield point being shifted to lower values. The variability in material behaviour and moisture content must be considered when predicting the block load. Moisture content variability lead to changing block bed load distributions. Underestimation of the moisture content can lead to large differences in the single dock block analysis. With the good matching in the validation work, this research confirms the reliability of a double Winkler system to prescribe non­linear material behaviour of a single dock block. Although the analysis of a single dock block matches results obtained from finite element models, further research is still needed. The model of a Timoshenko beam on spring supports gives reasonable results, but underestimates the load in case of local increased stiffness caused by a transverse bulkhead. Optimization on this particular section is needed to get more realistic results.

To reduce the environmental impact or cost of a civil engineering structure their designs are optimized. A promising method to optimize are iterative optimization algorithms. If both the design calculations and the iterative optimization algorithm are automated, an optimized design solution can be found within a feasible timeframe. To be able to automate the design calculations they need to be fully parameterizable. In this context it becomes interesting to research whether yet unparameterizable calculation processes can be made parameterizable. One of these processes is the determination of the locations in which the fatigue resistance has to be determined in a steel orthotropic bridge deck. This location is the location where the first fatigue crack is expected. Therefore, Antea Group requested if a study could be performed with the objective to answer the following research question: How can the determination of the location of the first fatigue crack in the deck, at a stiffener to deck plate weld toe, be parameterized? To answer the research question, the (in the Netherlands active) regulations are studied. Based on the regulations the process of determining fatigue damage of a point in the bridge can be understood. As well as the reason why, this process is too computational demanding and complex to be able to be applied to all points in all welds. In response to this an alternative method is proposed. This method reduces the complexity and the computational budget that is needed, by using 1D elements instead of the currently prescribed 2D elements. To determine if this method can be used it was decided to apply it on a case study. The bridge which served as the case study was the Goereese bridge. The alternative method was applied to determine the expected distribution of fatigue damages in all welds in the case study. Based on this obtained distribution a limited number of interesting locations in the deck could be identified. At these points to regulatory required method was used to obtain results which can be compared with the alternative method. It is concluded that the predicted location of the first fatigue crack of both methods is directly next to each other. However, the distribution of the remaining points suggest by the alternative method does not agree with the obtained results of the regulatory method. Remarkable enough, both these methods predict a location which is counter intuitive to the structural engineers participating in the research. Therefore, the following general recommendations are given:- Research if the regulatory method, to determine the location of the first fatigue crack, can be simplified. - Research the cause(s) of the differences between the regulatory method and the alternative method. - Increase the awareness of structural engineers regarding their intuition on the location of the first fatigue crack.

The slip joint connection is a relatively new alternative method for connecting offshore wind turbine towers to prior installed monopiles. It consists of two overlapping conical sections that together form a connection. There are several challenges left to be solved before the slip joint can be applied on a commercial scale. One of those challenges lies in the decommissioning process of the slip joint connection. It is proven that build-up settlement can increase the friction force within the joint over its lifetime and complicate its disconnection. One potential method for reducing the friction force in the slip joint is the excitation of one of the structures shell modes localized around the slip joint. Good estimations of the shell modes of the structure are therefore essential. Models for precise estimation of the shell modes of a wind turbine with a slip joint are difficult and expensive to develop due to the complexity of the joint. This thesis is a study into the possibility of using simplified finite element models to estimate the shell modes of a wind turbine connected with a slip joint. This is done by estimating the shell modes of the structure using a reference model and seven simplified models in a modal analysis. The estimated shell modes are compared based on their eigenfrequency and mode shape. Based on this comparison, conclusions about the applicability of simplified models are drawn. The first 300 eigenmodes of the structure are estimated with both the reference model and simplified models. Of these eigenmodes, ten shell modes of interest are selected based on their eigenfrequency and location at which the modes are localized. The modes of interest are than compared for their eigenfrequency and mode shape. The mode shapes are compared based on the Modal Assurance Criterion (MAC), which is calculated at a specific location of interest. In addition to this, the angle in which the modes are orientated is measured and compared, as this can be useful information for the decommissioning process. Results showed that three out of seven simplified models studied resulted in a estimation of the shell modes which was sufficiently accurate. The simplifications used for these models consist of the averaging of wall thickness for the upper and lower slip joint and a simple model for the upper slip joint. From these results it can be concluded that some of the simplifications can be applied when estimating shell modes of a wind turbine tower connected with a slip joint, as they lead to only a small decrease in accuracy of the shell mode estimation. However, using these results to make predictions about other potential simplifications is challenging as each different simplification has a unique influence on the estimation of the shell modes.

Friction damping is common in engineering structures for the purpose of energy dissipation and vibration control. Examples of applications are bolted connections and earthquake isolation systems. However, there is a shortage of works that investigate the effects of different contact materials on the energy dissipation performance and friction behaviour of friction dampers, which is valuable knowledge for design optimization. This thesis uses a numerical approach to explore the time response and energy dissipated by friction of the harmonically excited SDOF (single-degree-of-freedom) system with Coulomb friction contact between the sliding mass and a fixed wall. In addition, for the same SDOF system, an experimental investigation of the friction damping performance in terms of friction behaviour and energy dissipation is carried out for (1) steel, (2) rubber and (3) aramid contact materials. The aim is to get a better understanding of how different contact materials affect the performance of friction dampers. Various time scales, excitation frequencies and friction forces are considered. The main findings of this research are: (1) the characterization of the friction behaviour of steel-to-steel, rubber-to-steel and aramid-to-steel contacts; (2) the comparative analysis of the energy dissipation performance of the different contact materials; (3) the assessment of the long-term performance of the different contacts; (4) the comparison between numerical results based on the Coulomb friction model and experimental results. The tests have shown that rubber has the highest energy dissipation capacity and fairly unstable behaviour, steel has the second highest energy dissipation and irregular behaviour and aramid has the lowest energy dissipation performance and very consistent behaviour. Finally, the application of a method that calculates the energy dissipation of friction damping based on direct experimental outputs is an important contribution to the field regarding experimental investigations.

Bayesian system identification, including parameter estimation and model selection, is widely used to infer partially known, unobservable parameters of the models of physical systems when measurement data is available. A common assumption in the Bayesian system identification literature is that the discrepancy between model predictions and measurements can be described as independent, identically distributed realizations from a univariate Gaussian distribution. However, the decreasing cost of sensors and monitoring systems leads to more frequent structural measurements in close proximity to each other (e.g. fiber optics and strain gauges). In such cases, dependency in modeling uncertainty could be significant, both in space and time, and the assumption of uncorrelated Gaussian error may lead to inaccurate parameter estimation.The aim of this thesis is to explore how Bayesian system identification can be feasibly performed using large datasets when spatial and/or temporal dependence might be present and to assess the impact of considering this dependence. A pool of models, each assuming a different correlation structure, is defined and Bayesian inference is performed. In particular, stress measurements obtained on a steel road bridge are used to update the parameters of the corresponding FE model and the parameters of the correlation structure. The results are compared to a reference model where only measurements of the response peaks are used under the assumption of independence. Nested sampling is utilized to compute the evidence under each model and Bayesian model selection is applied. The question of efficiently performing system identification for large datasets (N > 102 for temporal dependencies and N > 103 for combined spatial and temporal dependencies) is investigated, and a novel approach for efficiently calculating the exact log-likelihood is derived. An approximation based on the Fisher information matrix is used to efficiently calculate the information content of measurements.It is found that the choice of correlation function can significantly affect the posterior distribution of the model prediction uncertainty. Additionally, it is shown that using large datasets and considering dependence makes it possible to perform system identification for a larger number of parameters compared to the reference model. The results of the case study indicate that using measurements from multiple sensors under combined spatial and temporal dependence and additive model prediction error yields reduced uncertainty in the posterior and up to 29% reduction of the posterior predictive credible interval range compared to the reference case. Furthermore, the efficiency of the proposed likelihood evaluation method is assessed. Using this method, exact calculation of the log-likelihood can be performed for >106 points in under a second in the case of correlation in one dimension. For combined spatial and temporal correlation it is shown to be approximately 900 times faster than naive evaluation for a 64 by 64 grid of observations. The results of the case study indicate that the described approach can be feasibly applied to real-world structures and can potentially improve parameter estimation and reduce prediction uncertainty. These findings suggest that further research into the approach could yield improvements over current methods.

An issue of utmost significance constitutes the maintenance of engineering systems exposed to corrosive environments, e.g. coastal and marine environments, highly acidic environments, etc. The most beneficial sequence of maintenance decisions, i.e. the one that corresponds to the minimum maintenance cost, can be sought as the solution to an optimization problem. Owing to the high complexity of this sequential decision optimization problem, traditional methods such as thresholdbased approaches, fail to arrive at an optimal strategy, while at the same time the commonly used offline knowledge about the environment can not capture efficiently the stochastic way in which an engineering system deteriorates. Over the last few years, Deep Reinforcement Learning (DRL) has been proven a promising tool to tackle such problems, being often limited though by the dimensionality curse and the implications caused by large state and action spaces, an issue which leads to simplifications like their discretization. Bayesian principles and model updating are the most widely used tools to model accurately systems with high uncertainty, by incorporating data acquired through monitoring devices and thus improving the knowledge about the stochastic system. This research proposes an integrated framework that aims to determine an optimal sequence of maintenance decisions over the lifespan of deteriorating engineering systems, combining the aforementioned core concepts of Deep Reinforcement Learning (DRL) and Bayesian Model Updating (BMU). More specifically, it investigates different Deep Reinforcement Learning (DRL) algorithms, namely Double Deep Q-Network (DDQN), Advantage Actor Critic (A2C), and Proximal Policy Optimization (PPO), while the updating of the uncertain parameters is performed through sampling, i.e. No-U-Turn Sampler (NUTS). All these tools will be first applied to elementary problems for the sake of verification and validation, while the culmination of this research is the application of the framework on a more realistic and complicated, multi-component structure. The obtained results are compared with benchmark performances to properly showcase the efficiency and the weaknesses of the tool.